transforms.Resize((IMAGE_SIZE, IMAGE_SIZE)), transforms.CenterCrop(IMAGE_SIZE), transforms.ToTensor(), tensor_normalizer() ]) img_tensor = transform(img).unsqueeze(0) if torch.cuda.is_available(): img_tensor = img_tensor.cuda() img_output = transformer(Variable(img_tensor, volatile=True)) plt.imshow(recover_image(img_tensor.cpu().numpy())[0]) Image.fromarray(recover_image(img_output.data.cpu().numpy())[0]) save_model_path = "model_udnie_imagenet_resnet2.pth" torch.save(transformer.state_dict(), save_model_path) transformer.load_state_dict(torch.load(save_model_path)) img = Image.open("content_images/amber.jpg").convert('RGB') transform = transforms.Compose([transforms.ToTensor(), tensor_normalizer()]) img_tensor = transform(img).unsqueeze(0) print(img_tensor.size()) if torch.cuda.is_available(): img_tensor = img_tensor.cuda() img_output = transformer(Variable(img_tensor, volatile=True)) plt.imshow(recover_image(img_tensor.cpu().numpy())[0]) plt.imshow(recover_image(img_output.data.cpu().numpy())[0])
def train(**kwargs): opt = Config() for k_, v_ in kwargs.items(): setattr(opt, k_, v_) # 可视化操作 vis = utils.Visualizer(opt.env) # 数据加载 transfroms = tv.transforms.Compose([ # 将输入的`PIL.Image`重新改变大小成给定的`size` `size`是最小边的边长 tv.transforms.Scale(opt.image_size), tv.transforms.CenterCrop(opt.image_size), # 转为0-1之间 tv.transforms.ToTensor(), # 转为0-255之间 tv.transforms.Lambda(lambda x: x * 255) ]) # 封装数据集,并进行数据转化 dataset = tv.datasets.ImageFolder(opt.data_root, transfroms) # 数据加载器 dataloader = data.DataLoader(dataset, opt.batch_size) # 转换网络 transformer = TransformerNet() if opt.model_path: transformer.load_state_dict( t.load(opt.model_path, map_location=lambda _s, _: _s)) # 损失网络 Vgg16 置为预测模式 vgg = Vgg16().eval() # 优化器(需要训练 风格转化网络的参数) optimizer = t.optim.Adam(transformer.parameters(), opt.lr) # 获取风格图片的数据 形状 1*c*h*w, 分布 -2~2(使用预设) style = utils.get_style_data(opt.style_path) # 可视化风格图:-2 到2 转化为0-1 vis.img('style', (style[0] * 0.225 + 0.45).clamp(min=0, max=1)) if opt.use_gpu: transformer.cuda() style = style.cuda() vgg.cuda() # 风格图片的gram矩阵 style_v = Variable(style, volatile=True) # 得到vgg中间四层的结果(用以跟输入图片的输出四层比较,计算损失) features_style = vgg(style_v) # gram_matrix:输入 b,c,h,w 输出 b,c,c 计算gram矩阵(四层的gram矩阵) gram_style = [Variable(utils.gram_matrix(y.data)) for y in features_style] # 损失统计 仪表盘 用以可视化(每个epoch中的所有batch平均损失) # 风格损失 style_meter = tnt.meter.AverageValueMeter() # 内容损失 content_meter = tnt.meter.AverageValueMeter() for epoch in range(opt.epoches): # 仪表盘清零 content_meter.reset() style_meter.reset() for ii, (x, _) in tqdm.tqdm(enumerate(dataloader)): # 训练 optimizer.zero_grad() if opt.use_gpu: x = x.cuda() # x为输入的真实图像 x = Variable(x) # 风格转换后的预测图像为y y = transformer(x) # 输入: b, ch, h, w 0~255 # 输出: b, ch, h, w - 2~2 # 将x,y范围从0-255转化为-2-2 y = utils.normalize_batch(y) x = utils.normalize_batch(x) # 返回 四个中间层的特征输出 features_y = vgg(y) features_x = vgg(x) # content loss内容损失 只计算relu2_2之间的损失 预测图片与原图在relu2_2中间层比较,计算损失 # content_weight内容的权重 mse_loss均方误差损失函数 content_loss = opt.content_weight * F.mse_loss( features_y.relu2_2, features_x.relu2_2) # style loss style_loss = 0. # 风格损失取四层的均方误差损失总和 # features_y:预测图像的四层输出内容 gram_style:风格图像的四层输出的gram_matrix # zip将可迭代的对象作为参数,将对象中对应的元素打包成一个个元组,然后返回由这些元组组成的列表 for ft_y, gm_s in zip(features_y, gram_style): # 计算预测图像的四层输出内容的gram_matrix gram_y = utils.gram_matrix(ft_y) style_loss += F.mse_loss(gram_y, gm_s.expand_as(gram_y)) style_loss *= opt.style_weight # 总损失=风格损失+内容损失 total_loss = content_loss + style_loss # 反向传播 total_loss.backward() # 更新参数 optimizer.step() # 损失平滑 将损失加入仪表盘,以便可视化损失过程 content_meter.add(content_loss.data[0]) style_meter.add(style_loss.data[0]) # 每plot_every次前向传播后可视化 if (ii + 1) % opt.plot_every == 0: if os.path.exists(opt.debug_file): ipdb.set_trace() # 可视化 vis.plot('content_loss', content_meter.value()[0]) vis.plot('style_loss', style_meter.value()[0]) # 因为x和y经过标准化处理(utils.normalize_batch),所以需要将它们还原 #x,y为[-2,2]还原回[0,1] vis.img('output', (y.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) vis.img('input', (x.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) # 每次epoch完毕后保存visdom和模型 vis.save([opt.env]) t.save(transformer.state_dict(), 'checkpoints/%s_style.pth' % epoch)
def train(args): if torch.cuda.is_available(): print('CUDA available, using GPU.') device = torch.device('cuda') else: print('GPU training unavailable... using CPU.') device = torch.device('cpu') np.random.seed(args.seed) torch.manual_seed(args.seed) transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size) # Image transformation network. transformer = TransformerNet() if args.model: state_dict = torch.load(args.model) transformer.load_state_dict(state_dict) transformer.to(device) optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() # Loss Network: VGG16 vgg = Vgg16(requires_grad=False).to(device) style_transform = transforms.Compose([ transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) style = utils.load_image(args.style_image, size=args.style_size) style = style_transform(style) style = style.repeat(args.batch_size, 1, 1, 1).to(device) features_style = vgg(utils.normalize_batch(style)) gram_style = [utils.gram_matrix(y) for y in features_style] for e in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) count += n_batch optimizer.zero_grad() # CUDA if available x = x.to(device) # Transform image y = transformer(x) y = utils.normalize_batch(y) x = utils.normalize_batch(x) # Feature map of original image features_x = vgg(x) # Feature Map of transformed image features_y = vgg(y) # Difference between transformed image, original image. # Changed to pull from features_.relu3_3 vs .relu2_2 content_loss = args.content_weight * mse_loss(features_y.relu3_3, features_x.relu3_3) # Compute gram matrix (dot product across each dimension G(4,3) = F4 * F3) style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = utils.gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[:n_batch, :, :]) style_loss *= args.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.item() agg_style_loss += style_loss.item() if True: #(batch_id + 1) % args.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1) ) print(mesg) if args.checkpoint_model_dir is not None and (batch_id + 1) % args.checkpoint_interval == 0: transformer.eval().cpu() ckpt_model_filename = "ckpt_epoch_" + str(e) + "_batch_id_" + str(batch_id + 1) + ".pth" ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) torch.save(transformer.state_dict(), ckpt_model_path) transformer.to(device).train() # save model transformer.eval().cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str(time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(args): np.random.seed(args.seed) torch.manual_seed(args.seed) if args.cuda: torch.cuda.manual_seed(args.seed) kwargs = {'num_workers': 0, 'pin_memory': False} else: kwargs = {} class RGB2YUV(object): def __call__(self, img): import numpy as np import cv2 npimg = np.array(img) yuvnpimg = cv2.cvtColor(npimg, cv2.COLOR_RGB2YUV) pilimg = Image.fromarray(yuvnpimg) return pilimg transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), RGB2YUV(), transforms.ToTensor(), # transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size, **kwargs) transformer = TransformerNet(in_channels=1, out_channels=2) # input: Y, predict: UV optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() # vgg = Vgg16() # utils.init_vgg16(args.vgg_model_dir) # vgg.load_state_dict(torch.load(os.path.join(args.vgg_model_dir, "vgg16.weight"))) transformer = nn.DataParallel(transformer) if args.cuda: if not torch.cuda.is_available(): raise RuntimeError( "CUDA is requested, but related driver/device is not set properly." ) transformer.cuda() for e in range(args.epochs): transformer.train() # agg_content_loss = 0. # agg_style_loss = 0. count = 0 for batch_id, (imgs, _) in enumerate(train_loader): n_batch = len(imgs) count += n_batch optimizer.zero_grad() # First channel x = imgs[:, :1, :, :].clone() # Second and third channels gt = imgs[:, 1:, :, :].clone() if args.cuda: x = x.cuda() gt = gt.cuda() y = transformer(x) total_loss = mse_loss(y, gt) total_loss.backward() optimizer.step() if (batch_id + 1) % args.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), total_loss / (batch_id + 1)) print(mesg) # save model transformer.eval() transformer.cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str( time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" os.makedirs(args.save_model_dir, exist_ok=True) save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
style_loss = 0. for m in range(len(features_y)): gram_s = gram_style[m] gram_y = gram_matrix(features_y[m]) style_loss += args.style_weight * loss(gram_y, gram_s.expand_as(gram_y)) total_loss = content_loss + style_loss + reg_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.data[0] agg_style_loss += style_loss.data[0] agg_reg_loss += reg_loss.data[0] if (batch_id + 1) % args.log_interval == 0: mesg = "[{}/{}] content: {:.6f} style: {:.6f} reg: {:.6f} total: {:.6f}".format( count, len(train_dataset), agg_content_loss / count, agg_style_loss / count, agg_reg_loss / count, (agg_content_loss + agg_style_loss + agg_reg_loss) / count) print(mesg) # save model transformer.eval() if torch.cuda.is_available(): transformer.cpu() model_file = 'model_' + str(epoch) + '.pth' torch.save(transformer.state_dict(), model_file) print('\nSaved model to ' + model_file + '.')
def fast_train(args): """Fast training""" device = torch.device("cuda" if args.cuda else "cpu") transformer = TransformerNet().to(device) if args.model: transformer.load_state_dict(torch.load(args.model)) vgg = Vgg16(requires_grad=False).to(device) global mse_loss mse_loss = torch.nn.MSELoss() content_weight = args.content_weight style_weight = args.style_weight lr = args.lr content_transform = transforms.Compose([ transforms.Resize(args.content_size), transforms.CenterCrop(args.content_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255))]) content_dataset = datasets.ImageFolder(args.content_dataset, content_transform) content_loader = DataLoader(content_dataset, batch_size=args.iter_batch_size, sampler=InfiniteSamplerWrapper(content_dataset), num_workers=args.n_workers) content_loader = iter(content_loader) style_transform = transforms.Compose([ transforms.Resize((args.style_size, args.style_size)), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255))]) style_image = utils.load_image(args.style_image) style_image = style_transform(style_image) style_image = style_image.unsqueeze(0).to(device) features_style = vgg(utils.normalize_batch(style_image.repeat(args.iter_batch_size, 1, 1, 1))) gram_style = [utils.gram_matrix(y) for y in features_style] if args.only_in: optimizer = Adam([param for (name, param) in transformer.named_parameters() if "in" in name], lr=lr) else: optimizer = Adam(transformer.parameters(), lr=lr) for i in trange(args.update_step): contents = content_loader.next()[0].to(device) features_contents = vgg(utils.normalize_batch(contents)) transformed = transformer(contents) features_transformed = vgg(utils.standardize_batch(transformed)) loss, c_loss, s_loss = loss_fn(features_transformed, features_contents, gram_style, content_weight, style_weight) optimizer.zero_grad() loss.backward() optimizer.step() # save model transformer.eval().cpu() style_name = os.path.basename(args.style_image).split(".")[0] save_model_filename = style_name + ".pth" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path)
def train(**kwargs): opt = Config() for k_, v_ in kwargs.items(): setattr(opt, k_, v_) device = t.device('cuda') if opt.use_gpu else t.device('cpu') vis = utils.Visualizer(opt.env) # 数据加载 transfroms = tv.transforms.Compose([ tv.transforms.Resize(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), tv.transforms.Lambda(lambda x: x * 255) ]) dataset = tv.datasets.ImageFolder(opt.data_root, transfroms) dataloader = data.DataLoader(dataset, opt.batch_size) # 转换网络 transformer = TransformerNet() if opt.model_path: transformer.load_state_dict( t.load(opt.model_path, map_location=lambda _s, _: _s)) transformer.to(device) # 损失网络 Vgg16 vgg = Vgg16().eval() vgg.to(device) for param in vgg.parameters(): param.requires_grad = False # 优化器 optimizer = t.optim.Adam(transformer.parameters(), opt.lr) # 获取风格图片的数据 style = utils.get_style_data(opt.style_path) vis.img('style', (style.data[0] * 0.225 + 0.45).clamp(min=0, max=1)) style = style.to(device) # 风格图片的gram矩阵 with t.no_grad(): features_style = vgg(style) gram_style = [utils.gram_matrix(y) for y in features_style] # 损失统计 style_meter = tnt.meter.AverageValueMeter() content_meter = tnt.meter.AverageValueMeter() for epoch in range(opt.epoches): content_meter.reset() style_meter.reset() for ii, (x, _) in tqdm.tqdm(enumerate(dataloader)): # 训练 optimizer.zero_grad() x = x.to(device) y = transformer(x) y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y) features_x = vgg(x) # content loss content_loss = opt.content_weight * F.mse_loss( features_y.relu2_2, features_x.relu2_2) # style loss style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gram_y = utils.gram_matrix(ft_y) style_loss += F.mse_loss(gram_y, gm_s.expand_as(gram_y)) style_loss *= opt.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() # 损失平滑 content_meter.add(content_loss.item()) style_meter.add(style_loss.item()) if (ii + 1) % opt.plot_every == 0: if os.path.exists(opt.debug_file): ipdb.set_trace() # 可视化 vis.plot('content_loss', content_meter.value()[0]) vis.plot('style_loss', style_meter.value()[0]) # 因为x和y经过标准化处理(utils.normalize_batch),所以需要将它们还原 vis.img('output', (y.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) vis.img('input', (x.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) # 保存visdom和模型 vis.save([opt.env]) t.save(transformer.state_dict(), 'checkpoints/%s_style.pth' % epoch)
def train(args): np.random.seed(args.seed) torch.manual_seed(args.seed) if args.cuda: torch.cuda.manual_seed(args.seed) kwargs = {'num_workers': 0, 'pin_memory': False} else: kwargs = {} training_set = np.loadtxt(args.dataset, dtype=np.float32) training_set_size = training_set.shape[1] num_batch = int(training_set_size / args.batch_size) transformer = TransformerNet() optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() vgg = Vgg16() utils.init_vgg16(args.vgg_model_dir) vgg.load_state_dict( torch.load(os.path.join(args.vgg_model_dir, "vgg16.weight"))) if args.cuda: transformer.cuda() vgg.cuda() style = np.loadtxt(args.style_image, dtype=np.float32) style = style.reshape((1, 1, args.style_size_x, args.style_size_y)) style = torch.from_numpy(style) style = style.repeat(args.batch_size, 3, 1, 1) if args.cuda: style = style.cuda() style_v = Variable(style, volatile=True) style_v = utils.subtract_imagenet_mean_batch(style_v) features_style = vgg(style_v) gram_style = [utils.gram_matrix(y) for y in features_style] # Hard data if args.hard_data: hard_data = np.loadtxt(args.hard_data_file) # if not isinstance(hard_data[0], list): # hard_data = [hard_data] for e in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 # for batch_id, (x, _) in enumerate(train_loader): for batch_id in range(num_batch): x = training_set[:, batch_id * args.batch_size:(batch_id + 1) * args.batch_size] n_batch = x.shape[1] count += n_batch x = x.transpose() x = x.reshape((n_batch, 1, args.image_size_x, args.image_size_y)) # plt.imshow(x[0,:,:,:].squeeze(0)) # plt.show() x = torch.from_numpy(x).float() optimizer.zero_grad() x = Variable(x) if args.cuda: x = x.cuda() y = transformer(x) if args.hard_data: hard_data_loss = 0 num_hard_data = 0 for hd in hard_data: hard_data_loss += args.hard_data_weight * ( y[:, 0, hd[1], hd[0]] - hd[2] * 255.0).norm()**2 / n_batch num_hard_data += 1 hard_data_loss /= num_hard_data y = y.repeat(1, 3, 1, 1) # x = Variable(utils.preprocess_batch(x)) # xc = x.data.clone() # xc = xc.repeat(1, 3, 1, 1) # xc = Variable(xc, volatile=True) y = utils.subtract_imagenet_mean_batch(y) # xc = utils.subtract_imagenet_mean_batch(xc) features_y = vgg(y) # features_xc = vgg(xc) # f_xc_c = Variable(features_xc[1].data, requires_grad=False) # content_loss = args.content_weight * mse_loss(features_y[1], f_xc_c) style_loss = 0. for m in range(len(features_y)): gram_s = Variable(gram_style[m].data, requires_grad=False) gram_y = utils.gram_matrix(features_y[m]) style_loss += args.style_weight * mse_loss( gram_y, gram_s[:n_batch, :, :]) # total_loss = content_loss + style_loss total_loss = style_loss if args.hard_data: total_loss += hard_data_loss total_loss.backward() optimizer.step() # agg_content_loss += content_loss.data[0] agg_style_loss += style_loss.data[0] if (batch_id + 1) % args.log_interval == 0: if args.hard_data: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\thard_data: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, num_batch, agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), hard_data_loss.data[0], (agg_content_loss + agg_style_loss) / (batch_id + 1)) else: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, num_batch, agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1)) print(mesg) # save model transformer.eval() transformer.cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str( time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def main(args): device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # DATA # Transform and Dataloader for COCO dataset transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), # / 255. transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size) # MODEL # Define Image Transformation Network with MSE loss and Adam optimizer transformer = TransformerNet().to(device) mse_loss = nn.MSELoss() optimizer = optim.Adam(transformer.parameters(), args.learning_rate) # Pretrained VGG vgg = VGG16(requires_grad=False).to(device) # FEATURES style_transform = transforms.Compose( [transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255))]) # Load the style image style = Image.open(args.style) style = style_transform(style) style = style.repeat(args.batch_size, 1, 1, 1).to(device) # Compute the style features features_style = vgg(normalize_batch(style)) # Loop through VGG style layers to calculate Gram Matrix gram_style = [gram_matrix(y) for y in features_style] # TRAIN for epoch in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. for batch_id, (x, _) in tqdm(enumerate(train_loader), unit='batch'): x = x.to(device) n_batch = len(x) optimizer.zero_grad() # Parse throught Image Transformation network y = transformer(x) y = normalize_batch(y) x = normalize_batch(x) # Parse through VGG layers features_y = vgg(y) features_x = vgg(x) # Calculate content loss content_loss = args.content_weight * mse_loss( features_y.relu2_2, features_x.relu2_2) # Calculate style loss style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[:n_batch, :, :]) style_loss *= args.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.item() agg_style_loss += style_loss.item() # Monitor if (batch_id + 1) % args.log_interval == 0: tqdm.write('[{}] ({})\t' 'content: {:.6f}\t' 'style: {:.6f}\t' 'total: {:.6f}'.format( epoch + 1, batch_id + 1, agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1))) # Checkpoint if (batch_id + 1) % args.save_interval == 0: # eval mode transformer.eval().cpu() style_name = args.style.split('/')[-1].split('.')[0] checkpoint_file = os.path.join(args.checkpoint_dir, '{}.pth'.format(style_name)) tqdm.write('Checkpoint {}'.format(checkpoint_file)) torch.save(transformer.state_dict(), checkpoint_file) # back to train mode transformer.to(device).train()
def train(args): # make sure each time we train, if args.seed stays the same, then # the random number we get is same as last time we train. np.random.seed(args.seed) torch.manual_seed(args.seed) if args.cuda: torch.cuda.manual_seed(args.seed) transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) # 0-1 to 0-255 ]) # note the order: give where the images at; load the images and transform; give the batch size train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size) # TODO: in transformernet transformer = TransformerNet() optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() # TODO: relus in vgg16 vgg = Vgg16(requires_grad=False) style_transform = transforms.Compose([ transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) style = utils.load_image(args.style_image, size=args.style_size) # style2 = utils.load_image(args.style_image2, size=args.style_size) style = style_transform(style) # style2 = style_transform(style2) # repeat the style tensor 4 times style = style.repeat(args.batch_size, 1, 1, 1) # style2 = style2.repeat(args.batch_size, 1, 1, 1) if args.cuda: transformer.cuda() vgg.cuda() style = style.cuda() # style2 = style2.cuda() style_v = Variable(style) style_v = utils.normalize_batch(style_v) features_style = vgg(style_v) # style_v2 = Variable(style2) # style_v2 = utils.normalize_batch(style_v2) # features_style2 = vgg(style_v2) # to determine style loss, make use of gram matrix gram_style = [utils.gram_matrix(y) for y in features_style] # gram_style2 = [utils.gram_matrix(y) for y in features_style2] for e in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) count += n_batch optimizer.zero_grad() # pytorch accumulates gradients, making them zero for each minibatch x = Variable(x) if args.cuda: x = x.cuda() # forward pass y = transformer(x) # after transformer - y y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y) features_x = vgg(x) # TODO: mse_loss of which relu could be modified content_loss = args.content_weight * mse_loss(features_y.relu2_2, features_x.relu2_2) style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = utils.gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[:n_batch, :, :]) # style_loss += mse_loss(gm_y, gm_s2[:n_batch, :, :]) style_loss *= args.style_weight total_loss = content_loss + style_loss # backward pass total_loss.backward() # this simply computes the gradients for each learnable parameters # update weights optimizer.step() agg_content_loss += content_loss.data[0] agg_style_loss += style_loss.data[0] if (batch_id + 1) % args.log_interval == 0: msg = "Epoch "+str(e + 1)+" "+str(count)+"/"+str(len(train_dataset)) msg += " content loss : "+str(agg_content_loss / (batch_id + 1)) msg += " style loss : " +str(agg_style_loss / (batch_id + 1)) msg += " total loss : " +str((agg_content_loss + agg_style_loss) / (batch_id + 1)) print(msg) if args.checkpoint_model_dir is not None and (batch_id + 1) % args.checkpoint_interval == 0: transformer.eval() if args.cuda: transformer.cpu() ckpt_model_filename = "ckpt_epoch_" + str(e) + "_batch_id_" + str(batch_id + 1) + ".pth" ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) torch.save(transformer.state_dict(), ckpt_model_path) if args.cuda: transformer.cuda() transformer.train() # save model transformer.eval() if args.cuda: transformer.cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str(time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(args): np.random.seed(args.seed) torch.manual_seed(args.seed) torch.cuda.manual_seed(args.seed) kwargs = {'num_workers': 0, 'pin_memory': False} transform = transforms.Compose([ transforms.Resize((args.image_size, args.image_size)), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = dataset.CustomImageDataset(args.dataset, transform=transform, img_size=args.image_size) train_loader = DataLoader(train_dataset, batch_size=args.batch_size, **kwargs) transformer = TransformerNet(args.pad_type) transformer = transformer.train() optimizer = torch.optim.Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() #print(transformer) vgg = Vgg16() vgg.load_state_dict( torch.load(os.path.join(args.vgg_model_dir, "vgg16.weight"))) vgg.eval() transformer = transformer.cuda() vgg = vgg.cuda() style = utils.tensor_load_resize(args.style_image, args.style_size) style = style.unsqueeze(0) print("=> Style image size: " + str(style.size())) #(1, H, W, C) style = utils.preprocess_batch(style).cuda() utils.tensor_save_bgrimage( style[0].detach(), os.path.join(args.save_model_dir, 'train_style.jpg'), True) style = utils.subtract_imagenet_mean_batch(style) features_style = vgg(style) gram_style = [utils.gram_matrix(y).detach() for y in features_style] for e in range(args.epochs): train_loader.dataset.reset() agg_content_loss = 0. agg_style_loss = 0. iters = 0 for batch_id, (x, _) in enumerate(train_loader): if x.size(0) != args.batch_size: print("=> Skip incomplete batch") continue iters += 1 optimizer.zero_grad() x = utils.preprocess_batch(x).cuda() y = transformer(x) if (batch_id + 1) % 1000 == 0: idx = (batch_id + 1) // 1000 utils.tensor_save_bgrimage( y.data[0], os.path.join(args.save_model_dir, "out_%d.png" % idx), True) utils.tensor_save_bgrimage( x.data[0], os.path.join(args.save_model_dir, "in_%d.png" % idx), True) y = utils.subtract_imagenet_mean_batch(y) x = utils.subtract_imagenet_mean_batch(x) features_y = vgg(y) features_x = vgg(center_crop(x, y.size(2), y.size(3))) #content target f_x = features_x[2].detach() # content f_y = features_y[2] content_loss = args.content_weight * mse_loss(f_y, f_x) style_loss = 0. for m in range(len(features_y)): gram_s = gram_style[m] gram_y = utils.gram_matrix(features_y[m]) batch_style_loss = 0 for n in range(gram_y.shape[0]): batch_style_loss += args.style_weight * mse_loss( gram_y[n], gram_s[0]) style_loss += batch_style_loss / gram_y.shape[0] total_loss = content_loss + style_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.data agg_style_loss += style_loss.data mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, batch_id + 1, len(train_loader), agg_content_loss / iters, agg_style_loss / iters, (agg_content_loss + agg_style_loss) / iters) print(mesg) agg_content_loss = agg_style_loss = 0.0 iters = 0 # save model save_model_filename = "epoch_" + str(e) + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(args): if not os.path.exists(args.save_model_dir): os.makedirs(args.save_model_dir) device = torch.device("cuda" if args.is_cuda else "cpu") np.random.seed(args.seed) torch.manual_seed(args.seed) transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size) transformer = TransformerNet().to(device) optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() vgg = Vgg16(requires_grad=False).to(device) style_transform = transforms.Compose([ transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) style = utils.load_image(args.style_image, size=args.style_size) # print(style.size) # ss('yo') style = style_transform(style) # it's not transform style = style.repeat(args.batch_size, 1, 1, 1).to(device) # style = style.repeat(2,1,1,1).to(device) # print(style.shape) # print() # ss('ho') features_style = vgg(utils.normalize_batch(style)) # print(features_style.relu4_3.shape) # for i in features_style: # print(i.shape) # ss('normalize') gram_style = [utils.gram_matrix(y) for y in features_style] # for i in gram_style: # print(i.shape) # ss('main: gram style') for e in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) # print(n_batch) # ss('hi') count += n_batch optimizer.zero_grad() x = x.to(device) # print(x.shape) # print(x[0,0,0,:]) # ss('in epoch, batch') y = transformer(x) # ss('in epoch, batch') y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y) features_x = vgg(x) content_loss = args.content_weight * mse_loss(features_y.relu2_2, features_x.relu2_2) style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = utils.gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[:n_batch, :, :]) style_loss *= args.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.item() agg_style_loss += style_loss.item() # if (batch_id + 1) % args.log_interval == 0: if True: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1) ) print(mesg) if args.is_quickrun: if count > 10: break # if args.checkpoint_model_dir is not None and (batch_id + 1) % args.checkpoint_interval == 0: # transformer.eval().cpu() # ckpt_model_filename = "ckpt_epoch_" + str(e) + "_batch_id_" + str(batch_id + 1) + ".pth" # ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) # torch.save(transformer.state_dict(), ckpt_model_path) # transformer.to(device).train() if (e % 50 == 0) or (e>400 and e % 10 ==0): # utils.save_image(args.save_model_dir+'/imgs/npepoch_{}.png'.format(e), y[0].detach().cpu()) # torchvision.utils.save_image(y, './imgs/epoch_{}.png'.format(e), normalize=True) torchvision.utils.save_image(y, './imgs/before/epoch_{}.png'.format(e), normalize=True) y = y.clamp(0, 255) torchvision.utils.save_image(y, './imgs/non/epoch_{}.png'.format(e)) torchvision.utils.save_image(y, './imgs/after/epoch_{}.png'.format(e), normalize=True) # ss('yo') # save model transformer.eval().cpu() save_model_filename = "style_"+args.style_name+"_epoch_" + str(args.epochs) + "_" + str(time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(args): np.random.seed(args.seed) torch.manual_seed(args.seed) if args.cuda: torch.cuda.manual_seed(args.seed) print("Loading data") transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size) print "Building the model" transformer = TransformerNet() optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() vgg = Vgg16(requires_grad=False) style_transform = transforms.Compose([ transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) style = utils.load_image(args.style_image, size=args.style_size) style = style_transform(style) style = style.repeat(args.batch_size, 1, 1, 1) if args.cuda: transformer.cuda() vgg.cuda() style = style.cuda() style_v = Variable(style) style_v = utils.normalize_batch(style_v) features_style = vgg(style_v) gram_style = [utils.gram_matrix(y) for y in features_style] def multiply(loss, weight): return loss * weight def add(loss1, loss2): return loss1 + loss2 metrics_names = ['Content Loss', 'Style Loss', 'Total Loss'] with missinglink_project.create_experiment( transformer, display_name='Style Transfer PyTorch', optimizer=optimizer, train_data_object=train_loader, metrics={metrics_names[0]: multiply, metrics_names[1]: multiply, metrics_names[2]: add} ) as experiment: (wrapped_content_loss, wrapped_style_loss, wrapped_total_loss) = [experiment.metrics[metric_name] for metric_name in metrics_names] print("Starting to train") for e in experiment.epoch_loop(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in experiment.batch_loop(iterable=train_loader): n_batch = len(x) count += n_batch optimizer.zero_grad() x = Variable(x) if args.cuda: x = x.cuda() y = transformer(x) y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y) features_x = vgg(x) content_loss = mse_loss(features_y.relu2_2, features_x.relu2_2) content_loss = wrapped_content_loss(content_loss, args.content_weight) style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = utils.gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[:n_batch, :, :]) style_loss = wrapped_style_loss(style_loss, args.style_weight) total_loss = wrapped_total_loss(content_loss, style_loss) total_loss.backward() optimizer.step() agg_content_loss += content_loss.data[0] agg_style_loss += style_loss.data[0] if (batch_id + 1) % args.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1) ) print(mesg) if args.checkpoint_model_dir is not None and (batch_id + 1) % args.checkpoint_interval == 0: transformer.eval() if args.cuda: transformer.cpu() ckpt_model_filename = "ckpt_epoch_" + str(e) + "_batch_id_" + str(batch_id + 1) + ".pth" ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) torch.save(transformer.state_dict(), ckpt_model_path) if args.cuda: transformer.cuda() transformer.train() # save model transformer.eval() if args.cuda: transformer.cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str(time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(**kwargs): for k_, v_ in kwargs.items(): setattr(opt, k_, v_) if opt.vis is True: from visualize import Visualizer vis = Visualizer(opt.env) transforms = tv.transforms.Compose([ tv.transforms.Resize(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), #change value to (0,1) tv.transforms.Lambda(lambda x: x * 255) ]) #change value to (0,255) dataset = tv.datasets.ImageFolder(opt.data_root, transforms) dataloader = data.DataLoader(dataset, opt.batch_size) #value is (0,255) transformer = TransformerNet() if opt.model_path: transformer.load_state_dict( t.load(opt.model_path, map_location=lambda _s, _: _s)) vgg = VGG16().eval() for param in vgg.parameters(): param.requires_grad = False optimizer = t.optim.Adam(transformer.parameters(), opt.lr) style = utils.get_style_data(opt.style_path) vis.img('style', (style[0] * 0.225 + 0.45).clamp(min=0, max=1)) if opt.use_gpu: transformer.cuda() style = style.cuda() vgg.cuda() style_v = Variable(style.unsqueeze(0), volatile=True) features_style = vgg(style_v) gram_style = [Variable(utils.gram_matrix(y.data)) for y in features_style] style_meter = tnt.meter.AverageValueMeter() content_meter = tnt.meter.AverageValueMeter() for epoch in range(opt.epoches): content_meter.reset() style_meter.reset() for ii, (x, _) in tqdm.tqdm(enumerate(dataloader)): optimizer.zero_grad() if opt.use_gpu: x = x.cuda() #(0,255) x = Variable(x) y = transformer(x) #(0,255) y = utils.normalize_batch(y) #(-2,2) x = utils.normalize_batch(x) #(-2,2) features_y = vgg(y) features_x = vgg(x) #calculate the content loss: it's only used relu2_2 # i think should add more layer's result to calculate the result like: w1*relu2_2+w2*relu3_2+w3*relu3_3+w4*relu4_3 content_loss = opt.content_weight * F.mse_loss( features_y.relu2_2, features_x.relu2_2) content_meter.add(content_loss.data) style_loss = 0 for ft_y, gm_s in zip(features_y, gram_style): gram_y = utils.gram_matrix(ft_y) style_loss += F.mse_loss(gram_y, gm_s.expand_as(gram_y)) style_meter.add(style_loss.data) style_loss *= opt.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() if (ii + 1) % (opt.plot_every) == 0: if os.path.exists(opt.debug_file): ipdb.set_trace() vis.plot('content_loss', content_meter.value()[0]) vis.plot('style_loss', style_meter.value()[0]) vis.img('output', (y.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) vis.img('input', (x.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) vis.save([opt.env]) t.save(transformer.state_dict(), 'checkpoints/%s_style.pth' % epoch)
def train(DC): train_gpu_id = DC.train_gpu_id device = t.device('cuda', train_gpu_id) if DC.use_gpu else t.device('cpu') input_size = DC.input_size super_resol_factor = DC.super_resol_factor high_transforms = T.Compose([ T.Resize(input_size), T.CenterCrop(input_size), T.ToTensor(), T.Lambda(lambda x: x * 255) ]) low_transforms = T.Compose([ T.Resize(int(input_size / super_resol_factor)), T.CenterCrop(int(input_size / super_resol_factor)), T.ToTensor(), T.Lambda(lambda x: x * 255) ]) HighResol_dir = DC.HighResol_dir LowResol_dir = DC.LowResol_dir batch_size = DC.train_batch_size HighResol_data = ImageFolder(HighResol_dir, transform=high_transforms) LowResol_data = ImageFolder(LowResol_dir, transform=low_transforms) num_train_data = len(HighResol_data) HighResol_dataloader = t.utils.data.DataLoader(HighResol_data, batch_size=batch_size, shuffle=False, num_workers=DC.num_workers, drop_last=True) LowResol_dataloader = t.utils.data.DataLoader(LowResol_data, batch_size=batch_size, shuffle=False, num_workers=DC.num_workers, drop_last=True) # transform net transformer = TransformerNet() if DC.load_model: transformer.load_state_dict( t.load(DC.load_model, map_location=lambda storage, loc: storage)) transformer.to(device) # Loss net (vgg16) vgg = Vgg16().eval() vgg.to(device) for param in vgg.parameters(): param.requires_grad = False optimizer = t.optim.Adam(transformer.parameters(), DC.base_lr) # Start training train_imgs = 0 iteration = 0 for epoch in range(DC.max_epoch): for i, ((high_data, _), (low_data, _)) in tqdm.tqdm( enumerate(zip(HighResol_dataloader, LowResol_dataloader))): train_imgs += batch_size iteration += 1 optimizer.zero_grad() # Transformer net x = low_data.to(device) y = transformer(x) y = utils.normalize_batch(y) yc = high_data.to(device) yc = utils.normalize_batch(yc) features_y = vgg(y) features_yc = vgg(yc) # Content loss content_loss = DC.content_weight * \ nn.functional.mse_loss(features_y.relu2_2, features_yc.relu2_2) # content_loss = DC.content_weight * \ # nn.functional.mse_loss(features_y.relu3_3, # features_yc.relu3_3) content_loss.backward() optimizer.step() if iteration % DC.show_iter == 0: print('\nepoch: ', epoch) print('content loss: ', content_loss.data) print() if (epoch + 1) % 10 == 0: t.save(transformer.state_dict(), '{}_style.pth'.format(epoch))
def train(args): # 将torch.Tensor分配到的设备的对象CPU或GPU device = torch.device("cuda" if args.cuda else "cpu") # 初始化随机种子 np.random.seed(args.seed) # 为CPU设置种子用于生成随机数 torch.manual_seed(args.seed) """ 将多个transform组合起来使用 """ transform = transforms.Compose([ # 重新设定大小 transforms.Resize(args.image_size), # 将给定的Image进行中心切割 transforms.CenterCrop(args.image_size), # 把Image转成张量Tensor格式,大小范围为[0,1] transforms.ToTensor(), # 使用lambd作为转换器 transforms.Lambda(lambda x: x.mul(255)) ]) # 使用ImageFolder数据加载器,传入数据集的路径 # transform:一个函数,原始图片作为输入,返回一个转换后的图片 train_dataset = datasets.ImageFolder(args.dataset, transform) # 把上一步做成的数据集放入Data.DataLoader中,可以生成一个迭代器 # batch_size:int,每个batch加载多少样本 train_loader = DataLoader(train_dataset, batch_size=args.batch_size) # 加载模型TransformerNet到设备上 transformer = TransformerNet().to(device) # 我们选择Adam作为优化器 optimizer = Adam(transformer.parameters(), args.lr) # 均方损失函数 mse_loss = torch.nn.MSELoss() # 加载模型Vgg16到设备上 vgg = Vgg16(requires_grad=False).to(device) # 风格图片的处理 style_transform = transforms.Compose( [transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255))]) # 载入风格图片 style = utils.load_image(args.style_image, size=args.style_size) # 处理风格图片 style = style_transform(style) # repeat(*sizes)沿着指定的维度重复tensor style = style.repeat(args.batch_size, 1, 1, 1).to(device) # 特征风格归一化 features_style = vgg(utils.normalize_batch(style)) # 风格特征图计算Gram矩阵 gram_style = [utils.gram_matrix(y) for y in features_style] # 迭代训练 for e in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) count += n_batch # 把梯度置零,也就是把loss关于weight的导数变成0 optimizer.zero_grad() y = transformer(x.to(device)) y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y) features_x = vgg(x.cuda()) # 计算内容损失 content_loss = args.content_weight * mse_loss( features_y.relu2_2, features_x.relu2_2) # 计算风格损失 style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = utils.gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[:n_batch, :, :]) style_loss *= args.style_weight # 总损失 total_loss = content_loss + style_loss # 反向传播 total_loss.backward() # 更新参数 optimizer.step() agg_content_loss += content_loss.item() agg_style_loss += style_loss.item() # 准备打印相关信息,args.log_interval是最开头设置的好了的参数 if (batch_id + 1) % args.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1)) print(mesg) # 生成训练好的风格图片模型 and (batch_id + 1) % args.checkpoint_interval == 0 if args.checkpoint_model_dir is not None: transformer.eval().cpu() ckpt_model_filename = "ckpt_epoch_" + str( e) + "_batch_id_" + str(batch_id + 1) + ".pth" ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) torch.save(transformer.state_dict(), ckpt_model_path) transformer.to(device).train() # save model transformer.eval().cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(args): """Meta train the model""" device = torch.device("cuda" if args.cuda else "cpu") np.random.seed(args.seed) torch.manual_seed(args.seed) if args.cuda: torch.cuda.manual_seed(args.seed) # first move parameters to GPU transformer = TransformerNet().to(device) vgg = Vgg16(requires_grad=False).to(device) global optimizer optimizer = Adam(transformer.parameters(), args.meta_lr) global mse_loss mse_loss = torch.nn.MSELoss() content_loader, style_loader, query_loader = get_data_loader(args) content_weight = args.content_weight style_weight = args.style_weight lr = args.lr writer = SummaryWriter(args.log_dir) for iteration in trange(args.max_iter): transformer.train() # bookkeeping # using state_dict causes problems, use named_parameters instead all_meta_grads = [] avg_train_c_loss = 0.0 avg_train_s_loss = 0.0 avg_train_loss = 0.0 avg_eval_c_loss = 0.0 avg_eval_s_loss = 0.0 avg_eval_loss = 0.0 contents = content_loader.next()[0].to(device) features_contents = vgg(utils.normalize_batch(contents)) querys = query_loader.next()[0].to(device) features_querys = vgg(utils.normalize_batch(querys)) # learning rate scheduling lr = args.lr / (1.0 + iteration * 2.5e-5) meta_lr = args.meta_lr / (1.0 + iteration * 2.5e-5) for param_group in optimizer.param_groups: param_group['lr'] = meta_lr for i in range(args.meta_batch_size): # sample a style style = style_loader.next()[0].to(device) style = style.repeat(args.iter_batch_size, 1, 1, 1) features_style = vgg(utils.normalize_batch(style)) gram_style = [utils.gram_matrix(y) for y in features_style] fast_weights = OrderedDict((name, param) for (name, param) in transformer.named_parameters() if re.search(r'in\d+\.', name)) for j in range(args.meta_step): # run forward transformation on contents transformed = transformer(contents, fast_weights) # compute loss features_transformed = vgg(utils.standardize_batch(transformed)) loss, c_loss, s_loss = loss_fn(features_transformed, features_contents, gram_style, content_weight, style_weight) # compute grad grads = torch.autograd.grad(loss, fast_weights.values(), create_graph=True) # update fast weights fast_weights = OrderedDict((name, param - lr * grad) for ((name, param), grad) in zip(fast_weights.items(), grads)) avg_train_c_loss += c_loss.item() avg_train_s_loss += s_loss.item() avg_train_loss += loss.item() # run forward transformation on querys transformed = transformer(querys, fast_weights) # compute loss features_transformed = vgg(utils.standardize_batch(transformed)) loss, c_loss, s_loss = loss_fn(features_transformed, features_querys, gram_style, content_weight, style_weight) grads = torch.autograd.grad(loss / args.meta_batch_size, transformer.parameters()) all_meta_grads.append({name: g for ((name, _), g) in zip(transformer.named_parameters(), grads)}) avg_eval_c_loss += c_loss.item() avg_eval_s_loss += s_loss.item() avg_eval_loss += loss.item() writer.add_scalar("Avg_Train_C_Loss", avg_train_c_loss / args.meta_batch_size, iteration + 1) writer.add_scalar("Avg_Train_S_Loss", avg_train_s_loss / args.meta_batch_size, iteration + 1) writer.add_scalar("Avg_Train_Loss", avg_train_loss / args.meta_batch_size, iteration + 1) writer.add_scalar("Avg_Eval_C_Loss", avg_eval_c_loss / args.meta_batch_size, iteration + 1) writer.add_scalar("Avg_Eval_S_Loss", avg_eval_s_loss / args.meta_batch_size, iteration + 1) writer.add_scalar("Avg_Eval_Loss", avg_eval_loss / args.meta_batch_size, iteration + 1) # compute dummy loss to refresh buffer transformed = transformer(querys) features_transformed = vgg(utils.standardize_batch(transformed)) dummy_loss, _, _ = loss_fn(features_transformed, features_querys, gram_style, content_weight, style_weight) meta_updates(transformer, dummy_loss, all_meta_grads) if args.checkpoint_model_dir is not None and (iteration + 1) % args.checkpoint_interval == 0: transformer.eval().cpu() ckpt_model_filename = "iter_" + str(iteration + 1) + ".pth" ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) torch.save(transformer.state_dict(), ckpt_model_path) transformer.to(device).train() # save model transformer.eval().cpu() save_model_filename = "Final_iter_" + str(args.max_iter) + "_" + \ str(args.content_weight) + "_" + \ str(args.style_weight) + "_" + \ str(args.lr) + "_" + \ str(args.meta_lr) + "_" + \ str(args.meta_batch_size) + "_" + \ str(args.meta_step) + "_" + \ time.ctime() + ".pth" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print "Done, trained model saved at {}".format(save_model_path)
def train(args): np.random.seed(args.seed) torch.manual_seed(args.seed) torch.cuda.manual_seed(args.seed) # device = torch.device('cuda' if args.cuda and torch.cuda.is_available() else 'cpu') transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), # utils.RGB2LAB(), transforms.ToTensor(), # utils.LAB2Tensor(), ]) pert_transform = transforms.Compose([utils.ColorPerturb()]) trainset = utils.FlatImageFolder(args.dataset, transform, pert_transform) trainloader = DataLoader(trainset, batch_size=args.batch_size, shuffle=True, pin_memory=True, num_workers=4) model = TransformerNet() if args.gpus is not None: model = nn.DataParallel(model, device_ids=args.gpus) else: model = nn.DataParallel(model) if args.resume: state_dict = torch.load(args.resume) model.load_state_dict(state_dict) if args.cuda: model.cuda() optimizer = torch.optim.Adam(model.parameters(), args.lr) criterion = nn.MSELoss() start_time = datetime.now() for e in range(args.epochs): model.train() count = 0 acc_loss = 0.0 for batchi, (pert_img, ori_img) in enumerate(trainloader): count += len(pert_img) if args.cuda: pert_img = pert_img.cuda(non_blocking=True) ori_img = ori_img.cuda(non_blocking=True) optimizer.zero_grad() rec_img = model(pert_img) loss = criterion(rec_img, ori_img) loss.backward() optimizer.step() acc_loss += loss.item() if (batchi + 1) % args.log_interval == 0: mesg = '{}\tEpoch {}: [{}/{}]\ttotal loss: {:.6f}'.format( time.ctime(), e + 1, count, len(trainset), acc_loss / (args.log_interval)) print(mesg) acc_loss = 0.0 if args.checkpoint_dir and e + 1 != args.epochs: model.eval().cpu() ckpt_filename = 'ckpt_epoch_' + str(e + 1) + '.pth' ckpt_path = osp.join(args.checkpoint_dir, ckpt_filename) torch.save(model.state_dict(), ckpt_path) model.cuda().train() print('Checkpoint model at epoch %d saved' % (e + 1)) model.eval().cpu() if args.save_model_name: model_filename = args.save_model_name else: model_filename = "epoch_" + str(args.epochs) + "_" + str( time.ctime()).replace(' ', '_') + ".model" model_path = osp.join(args.save_model_dir, model_filename) torch.save(model.state_dict(), model_path) end_time = datetime.now() print('Finished training after %s, trained model saved at %s' % (end_time - start_time, model_path))
def run_train(args): np.random.seed(args.seed) torch.manual_seed(args.seed) print('running training process...') if args.semantic == 1: print( 'multilabels semantic feedforward neural style transfer training...' ) elif args.semantic == 0: print('normal feedforward neural style transfer training...') if args.semantic == 1: loss_net, content_losses, style_losses, content_masks, n_channels = train_preparation_mask( args) elif args.semantic == 0: loss_net, content_losses, style_losses, n_channels = train_preparation( args) if args.backend == 'cudnn': torch.backends.cudnn.enabled = True transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size) transform_net = TransformerNet(n_channels).to(device) mse_loss = nn.MSELoss() optimizer = optim.Adam(transform_net.parameters(), lr=args.learning_rate) iteration = [0] while iteration[0] <= args.epochs - 1: transform_net.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): stloss = 0. ctloss = 0. n_batch = len(x) count += n_batch optimizer.zero_grad() #stack color_content_masks into x as input x, x_ori = x.to(device), x.to(device).clone() x = preprocess(x) x_ori = preprocess(x_ori) if args.semantic == 1: x = torch.cat((x, content_masks), 1) y = transform_net(x) #compute pixel loss if args.semantic == 1: y_pix = torch.cat((y, content_masks), 1) elif args.semantic == 0: y_pix = y pixloss = 0. if args.pixel_weight > 0: pixloss = mse_loss(x, y_pix) * args.pixel_weight #compute content loss and style loss for ctl in content_losses: ctl.mode = 'capture' loss_net(x_ori) for ctl in content_losses: ctl.mode = 'loss' for stl in style_losses: stl.mode = 'loss' loss_net(y) for ctl in content_losses: ctloss += mse_loss(ctl.input, ctl.target) * args.content_weight if args.semantic == 1: for stl in style_losses: for u in range(len(stl.color_codes)): input_msk = stl.input_masks[u].expand_as(stl.input) input_masked = torch.mul(stl.input, input_msk) input_msk_mean = torch.mean(stl.input_masks[u]) input_local_G = gram_matrix(input_masked) if input_msk_mean > 0: input_local_G.div(stl.input.nelement() * input_msk_mean) loss_local = mse_loss(input_local_G, stl.target[u]) loss_local *= input_msk_mean #larger target areas multiples smaller style weight if input_msk_mean > 0.2: stloss += loss_local * args.style_weights[0] #smaller target areas multiples larger style weight elif input_msk_mean <= 0.2: #print('aaaaa') stloss += loss_local * args.style_weights[1] elif args.semantic == 0: for stl in style_losses: gram = gram_matrix(stl.input) stloss += mse_loss(gram, stl.target) * args.style_weights[0] loss = ctloss + stloss + pixloss loss.backward() optimizer.step() agg_content_loss += ctloss.item() agg_style_loss += stloss.item() if (batch_id + 1) % args.log_interval == 0: mesg = "{}, Epoch {}:\t[{}/{}], content: {:.6f}, style: {:.6f}, total: {:.6f}".format( time.ctime(), iteration[0], count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1)) print(mesg) if args.checkpoint_model_dir is not None and ( batch_id + 1) % args.checkpoint_interval == 0: transform_net.eval().cpu() ckpt_model_filename = "ckpt_epoch_" + str( iteration[0] + 1) + "_batch_id_" + str(batch_id + 1) + "_semantic_" + str( args.semantic) + ".pth" ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) torch.save(transform_net.state_dict(), ckpt_model_path) transform_net.to(device).train() iteration[0] += 1 #save final model transform_net.eval().cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str( time.ctime()).replace(' ', '_') + "_content_" + str( args.content_weight) + "_style_" + str( args.style_weights[0]) + "_semantic_" + str( args.semantic) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transform_net.state_dict(), save_model_path) print("\n training process is Done!, trained model saved at", save_model_path)
def train(start_epoch=0): np.random.seed(enums.seed) torch.manual_seed(enums.seed) if enums.cuda: torch.cuda.manual_seed(enums.seed) transform = transforms.Compose([ transforms.Resize(enums.image_size), transforms.CenterCrop(enums.image_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = datasets.ImageFolder(enums.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=enums.batch_size) transformer = TransformerNet() optimizer = Adam(transformer.parameters(), enums.lr) if enums.subcommand == 'resume': ckpt_state = torch.load(enums.checkpoint_model) transformer.load_state_dict(ckpt_state['state_dict']) start_epoch = ckpt_state['epoch'] optimizer.load_state_dict(ckpt_state['optimizer']) mse_loss = torch.nn.MSELoss() vgg = Vgg16(requires_grad=False) style_transform = transforms.Compose( [transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255))]) style = utils.load_image(enums.style_image, size=enums.style_size) style = style_transform(style) style = style.expand(enums.batch_size, *style.size()) # N,C,H,W if enums.cuda: transformer.cuda() vgg.cuda() style = style.cuda() style_v = Variable(style) style_v = utils.normalize_batch(style_v) features_style = vgg(style_v) gram_style = [utils.gram_matrix(y) for y in features_style] for e in range(start_epoch, enums.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) count += n_batch optimizer.zero_grad() x = Variable(x) if enums.cuda: x = x.cuda() y = transformer(x) y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y) features_x = vgg(x) content_loss = enums.content_weight * mse_loss( features_y.relu2_2, features_x.relu2_2) style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = utils.gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[:n_batch, :, :]) style_loss *= enums.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.data[0] agg_style_loss += style_loss.data[0] if (batch_id + 1) % enums.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1)) print(mesg) if enums.checkpoint_model_dir is not None and ( e + 1) % enums.checkpoint_interval == 0: # transformer.eval() if enums.cuda: transformer.cpu() ckpt_model_filename = "ckpt_epoch_" + str(e + 1) + ".pth" ckpt_model_path = os.path.join(enums.checkpoint_model_dir, ckpt_model_filename) save_checkpoint( { 'epoch': e + 1, 'state_dict': transformer.state_dict(), 'optimizer': optimizer.state_dict() }, ckpt_model_path) if enums.cuda: transformer.cuda() # transformer.train() # save model # transformer.eval() if enums.cuda: transformer.cpu() save_model_filename = "epoch_" + str(enums.epochs) + "_" + str( time.ctime()).replace(' ', '_') + "_" + str( enums.content_weight) + "_" + str(enums.style_weight) + ".model" save_model_path = os.path.join(enums.save_model_dir, save_model_filename) save_checkpoint( { 'epoch': e + 1, 'state_dict': transformer.state_dict(), 'optimizer': optimizer.state_dict() }, save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(**kwargs): opt = Config() for k_, v_ in kwargs.items(): setattr(opt, k_, v_) vis = utils.Visualizer(opt.env) # 数据加载 transfroms = tv.transforms.Compose([ tv.transforms.Scale(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), tv.transforms.Lambda(lambda x: x * 255) ]) dataset = tv.datasets.ImageFolder(opt.data_root, transfroms) dataloader = data.DataLoader(dataset, opt.batch_size) # 转换网络 transformer = TransformerNet() if opt.model_path: transformer.load_state_dict(t.load(opt.model_path, map_location=lambda _s, _: _s)) # 损失网络 Vgg16 vgg = Vgg16().eval() # 优化器 optimizer = t.optim.Adam(transformer.parameters(), opt.lr) # 获取风格图片的数据 style = utils.get_style_data(opt.style_path) vis.img('style', (style[0] * 0.225 + 0.45).clamp(min=0, max=1)) if opt.use_gpu: transformer.cuda() style = style.cuda() vgg.cuda() # 风格图片的gram矩阵 style_v = Variable(style, volatile=True) features_style = vgg(style_v) gram_style = [Variable(utils.gram_matrix(y.data)) for y in features_style] # 损失统计 style_meter = tnt.meter.AverageValueMeter() content_meter = tnt.meter.AverageValueMeter() for epoch in range(opt.epoches): content_meter.reset() style_meter.reset() for ii, (x, _) in tqdm.tqdm(enumerate(dataloader)): # 训练 optimizer.zero_grad() if opt.use_gpu: x = x.cuda() x = Variable(x) y = transformer(x) y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y) features_x = vgg(x) # content loss content_loss = opt.content_weight * F.mse_loss(features_y.relu2_2, features_x.relu2_2) # style loss style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gram_y = utils.gram_matrix(ft_y) style_loss += F.mse_loss(gram_y, gm_s.expand_as(gram_y)) style_loss *= opt.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() # 损失平滑 content_meter.add(content_loss.data[0]) style_meter.add(style_loss.data[0]) if (ii + 1) % opt.plot_every == 0: if os.path.exists(opt.debug_file): ipdb.set_trace() # 可视化 vis.plot('content_loss', content_meter.value()[0]) vis.plot('style_loss', style_meter.value()[0]) # 因为x和y经过标准化处理(utils.normalize_batch),所以需要将它们还原 vis.img('output', (y.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) vis.img('input', (x.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) # 保存visdom和模型 vis.save([opt.env]) t.save(transformer.state_dict(), 'checkpoints/%s_style.pth' % epoch)
def train(args): np.random.seed(args.seed) torch.manual_seed(args.seed) if args.cuda: torch.cuda.manual_seed(args.seed) kwargs = {'num_workers': 0, 'pin_memory': False} else: kwargs = {} transform = transforms.Compose([transforms.Scale(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255))]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size, **kwargs) transformer = TransformerNet() if (args.premodel != ""): transformer.load_state_dict(torch.load(args.premodel)) print("load pretrain model:"+args.premodel) optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() vgg = Vgg16() utils.init_vgg16(args.vgg_model_dir) vgg.load_state_dict(torch.load(os.path.join(args.vgg_model_dir, "vgg16.weight"))) if args.cuda: transformer.cuda() vgg.cuda() style = utils.tensor_load_rgbimage(args.style_image, size=args.style_size) style = style.repeat(args.batch_size, 1, 1, 1) style = utils.preprocess_batch(style) if args.cuda: style = style.cuda() style_v = Variable(style, volatile=True) style_v = utils.subtract_imagenet_mean_batch(style_v) features_style = vgg(style_v) gram_style = [utils.gram_matrix(y) for y in features_style] hori=0 writer = SummaryWriter(args.logdir,comment=args.logdir) for e in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. agg_cate_loss = 0. agg_cam_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) count += n_batch optimizer.zero_grad() x = Variable(utils.preprocess_batch(x)) if args.cuda: x = x.cuda() y = transformer(x) xc = Variable(x.data.clone(), volatile=True) #print(y.size()) #(4L, 3L, 224L, 224L) # Calculate focus loss and category loss y_cam = utils.depreprocess_batch(y) y_cam = utils.subtract_mean_std_batch(y_cam) xc_cam = utils.depreprocess_batch(xc) xc_cam = utils.subtract_mean_std_batch(xc_cam) del features_blobs[:] logit_x = net(xc_cam) logit_y = net(y_cam) label=[] cam_loss = 0 for i in range(len(xc_cam)): h_x = F.softmax(logit_x[i]) probs_x, idx_x = h_x.data.sort(0, True) label.append(idx_x[0]) h_y = F.softmax(logit_y[i]) probs_y, idx_y = h_y.data.sort(0, True) x_cam = returnCAM(features_blobs[0][i], weight_softmax, idx_x[0]) x_cam = Variable(x_cam.data,requires_grad = False) y_cam = returnCAM(features_blobs[1][i], weight_softmax, idx_y[0]) cam_loss += mse_loss(y_cam, x_cam) #the focus loss cam_loss *= 80 #the category loss label = Variable(torch.LongTensor(label),requires_grad = False).cuda() cate_loss = 10000 * torch.nn.CrossEntropyLoss()(logit_y,label) y = utils.subtract_imagenet_mean_batch(y) xc = utils.subtract_imagenet_mean_batch(xc) features_y = vgg(y) features_xc = vgg(xc) #f_xc_c = Variable(features_xc[1].data, requires_grad=False) #content_loss = args.content_weight * mse_loss(features_y[1], f_xc_c) f_xc_c = Variable(features_xc[2].data, requires_grad=False) content_loss = args.content_weight * mse_loss(features_y[2], f_xc_c) style_loss = 0. for m in range(len(features_y)): gram_s = Variable(gram_style[m].data, requires_grad=False) gram_y = utils.gram_matrix(features_y[m]) style_loss += args.style_weight * mse_loss(gram_y, gram_s[:n_batch, :, :]) #add the total four loss and backward total_loss = style_loss + content_loss + cam_loss + cate_loss total_loss.backward() optimizer.step() #something for display agg_content_loss += content_loss.data[0] agg_style_loss += style_loss.data[0] agg_cate_loss += cate_loss.data[0] agg_cam_loss += cam_loss.data[0] writer.add_scalar("Loss_Cont", agg_content_loss / (batch_id + 1), hori) writer.add_scalar("Loss_Style", agg_style_loss / (batch_id + 1), hori) writer.add_scalar("Loss_CAM", agg_cam_loss / (batch_id + 1), hori) writer.add_scalar("Loss_Cate", agg_cate_loss / (batch_id + 1), hori) hori += 1 if (batch_id + 1) % args.log_interval == 0: mesg = "{}Epoch{}:[{}/{}] content:{:.2f} style:{:.2f} cate:{:.2f} cam:{:.2f} total:{:.2f}".format( time.strftime("%a %H:%M:%S"),e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), agg_cate_loss / (batch_id + 1), agg_cam_loss / (batch_id + 1), (agg_content_loss + agg_style_loss + agg_cate_loss + agg_cam_loss ) / (batch_id + 1) ) print(mesg) if (batch_id + 1) % 2500 == 0: transformer.eval() transformer.cpu() save_model_filename = "epoch_" + str(e+1) + "_" + str(time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) transformer.cuda() transformer.train() print("saved at ",count) # save model transformer.eval() transformer.cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str(time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) writer.close() print("\nDone, trained model saved at", save_model_path)
def train(): train_gpu_id = DC.train_gpu_id device = t.device('cuda', train_gpu_id) if DC.use_gpu else t.device('cpu') transforms = T.Compose([ T.Resize(DC.input_size), T.CenterCrop(DC.input_size), T.ToTensor(), T.Lambda(lambda x: x*255) ]) train_dir = DC.train_content_dir batch_size = DC.train_batch_size train_data = ImageFolder(train_dir, transform=transforms) num_train_data = len(train_data) train_dataloader = t.utils.data.DataLoader(train_data, batch_size=batch_size, shuffle=True, num_workers=DC.num_workers, drop_last=True) # transform net transformer = TransformerNet() if DC.load_model: transformer.load_state_dict( t.load(DC.load_model, map_location=lambda storage, loc: storage)) transformer.to(device) # Loss net (vgg16) vgg = Vgg16().eval() vgg.to(device) for param in vgg.parameters(): param.requires_grad = False optimizer = t.optim.Adam(transformer.parameters(), DC.base_lr) # Get the data from style image ys = utils.get_style_data(DC.style_img) ys = ys.to(device) # The Gram matrix of the style image with t.no_grad(): features_ys = vgg(ys) gram_ys = [utils.gram_matrix(ys) for ys in features_ys] # Start training train_imgs = 0 iteration = 0 for epoch in range(DC.max_epoch): for i, (data, label) in tqdm.tqdm(enumerate(train_dataloader)): train_imgs += batch_size iteration += 1 optimizer.zero_grad() # Transformer net x = data.to(device) y = transformer(x) x = utils.normalize_batch(x) yc = x y = utils.normalize_batch(y) features_y = vgg(y) features_yc = vgg(yc) # Content loss content_loss = DC.content_weight * \ nn.functional.mse_loss(features_y.relu2_2, features_yc.relu2_2) # content_loss = DC.content_weight * \ # nn.functional.mse_loss(features_y.relu3_3, # features_yc.relu3_3) # Style loss style_loss = 0.0 for ft_y, gm_ys in zip(features_y, gram_ys): gm_y = utils.gram_matrix(ft_y) style_loss += nn.functional.mse_loss(gm_y, gm_ys.expand_as(gm_y)) style_loss *= DC.style_weight # Total loss total_loss = content_loss + style_loss total_loss.backward() optimizer.step() if iteration%DC.show_iter == 0: print('\ncontent loss: ', content_loss.data) print('style loss: ', style_loss.data) print('total loss: ', total_loss.data) print() t.save(transformer.state_dict(), '{}_style.pth'.format(epoch))
def train(args): np.random.seed(args.seed) torch.manual_seed(args.seed) if args.cuda: torch.cuda.manual_seed(args.seed) kwargs = {'num_workers': 12, 'pin_memory': False} else: kwargs = {} from transform.color_op import Linearize, SRGB2XYZ, XYZ2CIE RGB2YUV = transforms.Compose([ Linearize(), SRGB2XYZ(), XYZ2CIE() ]) transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), RGB2YUV(), transforms.ToTensor(), # transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size, **kwargs) transformer = TransformerNet(in_channels=2, out_channels=1) # input: LS, predict: M optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() transformer = nn.DataParallel(transformer) if args.cuda: if not torch.cuda.is_available(): raise RuntimeError("CUDA is requested, but related driver/device is not set properly.") transformer.cuda() for e in range(args.epochs): transformer.train() # agg_content_loss = 0. # agg_style_loss = 0. count = 0 for batch_id, (imgs, _) in enumerate(train_loader): n_batch = len(imgs) count += n_batch optimizer.zero_grad() # First channel x = torch.cat([imgs[:, :1, :, :].clone(), imgs[:, -1:, :, :].clone()], dim=1) # Second and third channels gt = imgs[:, 1:2, :, :].clone() if args.cuda: x = x.cuda() gt = gt.cuda() y = transformer(x) total_loss = mse_loss(y, gt) total_loss.backward() optimizer.step() if (batch_id + 1) % args.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), total_loss / (batch_id + 1) ) print(mesg) # save model transformer.eval() transformer.cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str(time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" os.makedirs(args.save_model_dir, exist_ok=True) save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(args): np.random.seed(args.seed) torch.manual_seed(args.seed) if args.cuda: torch.cuda.manual_seed(args.seed) kwargs = {'num_workers': 0, 'pin_memory': False} else: kwargs = {} transform = transforms.Compose([ transforms.Scale(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size, **kwargs) transformer = TransformerNet() optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() vgg = Vgg16() utils.init_vgg16(args.vgg_model_dir) vgg.load_state_dict( torch.load(os.path.join(args.vgg_model_dir, "vgg16.weight"))) if args.cuda: transformer.cuda() vgg.cuda() style = utils.tensor_load_rgbimage(args.style_image, size=args.style_size) style = style.repeat(args.batch_size, 1, 1, 1) style = utils.preprocess_batch(style) if args.cuda: style = style.cuda() style_v = Variable(style, volatile=True) utils.subtract_imagenet_mean_batch(style_v) features_style = vgg(style_v) gram_style = [utils.gram_matrix(y) for y in features_style] for e in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) count += n_batch optimizer.zero_grad() x = Variable(utils.preprocess_batch(x)) if args.cuda: x = x.cuda() y = transformer(x) xc = Variable(x.data.clone(), volatile=True) utils.subtract_imagenet_mean_batch(y) utils.subtract_imagenet_mean_batch(xc) features_y = vgg(y) features_xc = vgg(xc) f_xc_c = Variable(features_xc[1].data, requires_grad=False) content_loss = args.content_weight * mse_loss( features_y[1], f_xc_c) style_loss = 0. for m in range(len(features_y)): gram_s = Variable(gram_style[m].data, requires_grad=False) gram_y = utils.gram_matrix(features_y[m]) style_loss += args.style_weight * mse_loss( gram_y, gram_s[:n_batch, :, :]) total_loss = content_loss + style_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.data[0] agg_style_loss += style_loss.data[0] if (batch_id + 1) % args.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1)) print(mesg) # save model transformer.eval() transformer.cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str( time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(args): device = torch.device("cuda" if args.cuda else "cpu") np.random.seed(args.seed) torch.manual_seed(args.seed) transform = transforms.Compose([ transforms.Resize( args.image_size), # the shorter side is resize to match image_size transforms.CenterCrop(args.image_size), transforms.ToTensor(), # to tensor [0,1] transforms.Lambda(lambda x: x.mul(255)) # convert back to [0, 255] ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True) # to provide a batch loader style_image = [f for f in os.listdir(args.style_image)] style_num = len(style_image) print(style_num) transformer = TransformerNet(style_num=style_num).to(device) optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() vgg = Vgg16(requires_grad=False).to(device) style_transform = transforms.Compose([ transforms.Resize(args.style_size), transforms.CenterCrop(args.style_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) style_batch = [] for i in range(style_num): style = utils.load_image(args.style_image + style_image[i], size=args.style_size) style = style_transform(style) style_batch.append(style) style = torch.stack(style_batch).to(device) features_style = vgg(utils.normalize_batch(style)) gram_style = [utils.gram_matrix(y) for y in features_style] for e in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) if n_batch < args.batch_size: break # skip to next epoch when no enough images left in the last batch of current epoch count += n_batch optimizer.zero_grad() # initialize with zero gradients batch_style_id = [ i % style_num for i in range(count - n_batch, count) ] y = transformer(x.to(device), style_id=batch_style_id) y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y.to(device)) features_x = vgg(x.to(device)) content_loss = args.content_weight * mse_loss( features_y.relu2_2, features_x.relu2_2) style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = utils.gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[batch_style_id, :, :]) style_loss *= args.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.item() agg_style_loss += style_loss.item() if (batch_id + 1) % args.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1)) print(mesg) if args.checkpoint_model_dir is not None and ( batch_id + 1) % args.checkpoint_interval == 0: transformer.eval().cpu() ckpt_model_filename = "ckpt_epoch_" + str( e) + "_batch_id_" + str(batch_id + 1) + ".pth" ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) torch.save(transformer.state_dict(), ckpt_model_path) transformer.to(device).train() # save model transformer.eval().cpu() save_model_filename = "epoch_" + str( args.epochs) + "_" + str(time.ctime()).replace(' ', '_').replace( ':', '') + "_" + str(int(args.content_weight)) + "_" + str( int(args.style_weight)) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(**kwargs): for k_, v_ in kwargs.items(): setattr(opt, k_, v_) vis = utils.Visualizer(opt.env) # 数据加载 transform = transforms.Compose([ transforms.ToTensor(), transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)) ]) loader = get_loader(batch_size=1, data_path=opt.data_path, img_shape=opt.img_shape, transform=transform) # 转换网络 transformer = TransformerNet().cuda() # transformer.load_state_dict(t.load(opt.model_path, )) #if opt.model_path: # transformer.load_state_dict(t.load(opt.model_path,map_location=lambda _s, _: _s)) # 损失网络 Vgg16 vgg = Vgg19().eval() depthnet = HourGlass().eval() depthnet.load_state_dict(t.load(opt.depth_path)) # print(vgg) # BASNET net = BASNet(3, 1).cuda() net.load_state_dict(torch.load('./basnet.pth')) net.eval() # 优化器 optimizer = t.optim.Adam(transformer.parameters(), lr=opt.lr) # 获取风格图片的数据 img = Image.open(opt.style_path) img = img.resize(opt.img_shape) img = transform(img).float() style = Variable(img, requires_grad=True).unsqueeze(0) vis.img('style', (style[0] * 0.225 + 0.45).clamp(min=0, max=1)) if opt.use_gpu: transformer.cuda() style = style.cuda() vgg.cuda() depthnet.cuda() # 风格图片的gram矩阵 style_v = Variable(style, volatile=True) features_style = vgg(style_v) gram_style = [Variable(utils.gram_matrix(y.data)) for y in features_style] # 损失统计 style_meter = tnt.meter.AverageValueMeter() content_meter = tnt.meter.AverageValueMeter() temporal_meter = tnt.meter.AverageValueMeter() long_temporal_meter = tnt.meter.AverageValueMeter() depth_meter = tnt.meter.AverageValueMeter() # tv_meter = tnt.meter.AverageValueMeter() kk = 0 for count in range(opt.epoch): print('Training Start!!') content_meter.reset() style_meter.reset() temporal_meter.reset() long_temporal_meter.reset() depth_meter.reset() # tv_meter.reset() for step, frames in enumerate(loader): for i in tqdm.tqdm(range(1, len(frames))): kk += 1 if (kk + 1) % 3000 == 0: print('LR had changed') for param in optimizer.param_groups: param['lr'] = max(param['lr'] / 1.2, 1e-4) optimizer.zero_grad() x_t = frames[i].cuda() x_t1 = frames[i - 1].cuda() h_xt = transformer(x_t) h_xt1 = transformer(x_t1) depth_x_t = depthnet(x_t) depth_x_t1 = depthnet(x_t1) depth_h_xt = depthnet(h_xt) depth_h_xt1 = depthnet(h_xt1) img1 = h_xt1.data.cpu().squeeze(0).numpy().transpose(1, 2, 0) img2 = h_xt.data.cpu().squeeze(0).numpy().transpose(1, 2, 0) flow, mask = opticalflow(img1, img2) d1, d2, d3, d4, d5, d6, d7, d8 = net(x_t) a1pha1 = PROCESS(d1, x_t) del d1, d2, d3, d4, d5, d6, d7, d8 d1, d2, d3, d4, d5, d6, d7, d8 = net(x_t1) a1pha2 = PROCESS(d1, x_t1) del d1, d2, d3, d4, d5, d6, d7, d8 h_xt_features = vgg(h_xt) h_xt1_features = vgg(h_xt1) x_xt_features = vgg(a1pha1) x_xt1_features = vgg(a1pha2) # ContentLoss, conv3_2 content_t = F.mse_loss(x_xt_features[2], h_xt_features[2]) content_t1 = F.mse_loss(x_xt1_features[2], h_xt1_features[2]) content_loss = opt.content_weight * (content_t1 + content_t) # StyleLoss style_t = 0 style_t1 = 0 for ft_y, gm_s in zip(h_xt_features, gram_style): gram_y = gram_matrix(ft_y) style_t += F.mse_loss(gram_y, gm_s.expand_as(gram_y)) for ft_y, gm_s in zip(h_xt1_features, gram_style): gram_y = gram_matrix(ft_y) style_t1 += F.mse_loss(gram_y, gm_s.expand_as(gram_y)) style_loss = opt.style_weight * (style_t1 + style_t) # # depth loss depth_loss1 = F.mse_loss(depth_h_xt, depth_x_t) depth_loss2 = F.mse_loss(depth_h_xt1, depth_x_t1) depth_loss = opt.depth_weight * (depth_loss1 + depth_loss2) # # TVLoss # print(type(s_hxt[layer]),s_hxt[layer].size()) # tv_loss = TVLoss(h_xt) #Long-temprol loss if (i - 1) % opt.sample_frames == 0: frames0 = h_xt1.cpu() long_img1 = frames0.data.cpu().squeeze( 0).numpy().transpose(1, 2, 0) # long_img2 = h_xt.data.cpu().squeeze(0).numpy().transpose(1,2,0) long_flow, long_mask = opticalflow(long_img1, img2) # Optical flow flow = torch.from_numpy(flow).permute(2, 0, 1).unsqueeze(0).to( torch.float32) long_flow = torch.from_numpy(long_flow).permute( 2, 0, 1).unsqueeze(0).to(torch.float32) # print(flow.size()) # print(h_xt1.size()) warped = warp(h_xt1.cpu().permute(0, 2, 3, 1), flow, opt.img_shape[1], opt.img_shape[0]).cuda() long_warped = warp(frames0.cpu().permute(0, 2, 3, 1), long_flow, opt.img_shape[1], opt.img_shape[0]).cuda() long_temporal_loss = F.mse_loss( h_xt, long_mask * long_warped.permute(0, 3, 1, 2)) # print(warped.size()) # tv.utils.save_image((warped.permute(0,3,1,2).data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1), # './warped.jpg') mask = mask.transpose(2, 0, 1) mask = torch.from_numpy(mask).cuda().to(torch.float32) # print(mask.shape) temporal_loss = F.mse_loss(h_xt, mask * warped.permute(0, 3, 1, 2)) temporal_loss = opt.temporal_weight * temporal_loss long_temporal_loss = opt.long_temporal_weight * long_temporal_loss # Spatial Loss spatial_loss = content_loss + style_loss Loss = spatial_loss + depth_loss + temporal_loss + long_temporal_loss Loss.backward(retain_graph=True) optimizer.step() content_meter.add(float(content_loss.data)) style_meter.add(float(style_loss.data)) temporal_meter.add(float(temporal_loss.data)) long_temporal_meter.add(float(long_temporal_loss.data)) depth_meter.add(float(depth_loss.data)) # tv_meter.add(float(tv_loss.data)) vis.plot('temporal_loss', temporal_meter.value()[0]) vis.plot('long_temporal_loss', long_temporal_meter.value()[0]) vis.plot('content_loss', content_meter.value()[0]) vis.plot('style_loss', style_meter.value()[0]) vis.plot('depth_loss', depth_meter.value()[0]) # vis.plot('tv_loss', tv_meter.value()[0]) if i % 10 == 0: vis.img('input(t)', (x_t.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) vis.img('output(t)', (h_xt.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) vis.img('output(t-1)', (h_xt1.data.cpu()[0] * 0.225 + 0.45).clamp(min=0, max=1)) print( 'epoch{},content loss:{},style loss:{},temporal loss:{},long temporal loss:{},depth loss:{},total loss{}' .format(count, content_loss, style_loss, temporal_loss, long_temporal_loss, depth_loss, Loss)) # print('epoch{},content loss:{},style loss:{},depth loss:{},total loss{}' # .format(count,content_loss, style_loss,depth_loss,Loss)) vis.save([opt.env]) torch.save(transformer.state_dict(), opt.model_path)
def train(args): device = torch.device("cuda" if args.cuda else "cpu") np.random.seed(args.seed) torch.manual_seed(args.seed) transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) target_transform = transforms.ToTensor() train_dataset = VFDataset(args.dataset, transform, target_transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size) transformer = TransformerNet().to(device) if args.load_model is not None: transformer.load_state_dict(torch.load(args.load_model)) optimizer = Adam(transformer.parameters(), args.lr) # mse_loss = torch.nn.MSELoss() cosine_loss = torch.nn.CosineEmbeddingLoss() label = torch.ones(args.batch_size, 1, args.image_size, args.image_size).to(device) # log_file = open(args.log_file, "w") for e in range(args.epochs): transformer.train() agg_loss = 0. count = 0 for batch_id, (x, vf) in enumerate(train_loader): n_batch = len(x) count += n_batch optimizer.zero_grad() x = utils.subtract_imagenet_mean_batch(x) x = x.to(device) y = transformer(x) vf = vf.to(device) # loss = mse_loss(y, vf) loss = cosine_loss(y, vf, label) loss.backward() optimizer.step() agg_loss += loss.item() if (batch_id + 1) % args.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_loss / (batch_id + 1)) print(mesg) if args.checkpoint_model_dir is not None and ( batch_id + 1) % args.checkpoint_interval == 0: transformer.eval().cpu() ckpt_model_filename = "ckpt_epoch_" + str( e) + "_batch_id_" + str(batch_id + 1) + ".pth" ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) torch.save(transformer.state_dict(), ckpt_model_path) transformer.to(device).train() # save model transformer.eval().cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str( time.ctime()).replace(' ', '_') + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(args): device = torch.device("cuda" if args.cuda else "cpu") np.random.seed(args.seed) torch.manual_seed(args.seed) transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size) transformer = TransformerNet().to(device) optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() vgg = Vgg16(requires_grad=False).to(device) style_transform = transforms.Compose([ transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) style = utils.load_image(args.style_image, size=args.style_size) style = style_transform(style) style = style.repeat(args.batch_size, 1, 1, 1).to(device) features_style = vgg(utils.normalize_batch(style)) gram_style = [utils.gram_matrix(y) for y in features_style] for e in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) count += n_batch optimizer.zero_grad() x = x.to(device) y = transformer(x) y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y) features_x = vgg(x) content_loss = args.content_weight * mse_loss(features_y.relu2_2, features_x.relu2_2) style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = utils.gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[:n_batch, :, :]) style_loss *= args.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.item() agg_style_loss += style_loss.item() if (batch_id + 1) % args.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1) ) print(mesg) if args.checkpoint_model_dir is not None and (batch_id + 1) % args.checkpoint_interval == 0: transformer.eval().cpu() ckpt_model_filename = "ckpt_epoch_" + str(e) + "_batch_id_" + str(batch_id + 1) + ".pth" ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) torch.save(transformer.state_dict(), ckpt_model_path) transformer.to(device).train() # save model transformer.eval().cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str(time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(): device = torch.device("cuda") np.random.seed(random_seed) torch.manual_seed(random_seed) transform = transforms.Compose([ transforms.Resize(image_size), transforms.CenterCrop(image_size), transforms.ToTensor(), ]) train_dataset = datasets.ImageFolder(dataset_path, transform) train_loader = DataLoader(train_dataset, batch_size=batch_size) transformer = TransformerNet().to(device) optimizer = Adam(transformer.parameters(), lr) mse_loss = torch.nn.MSELoss() if resume_TransformerNet_from_file: if os.path.isfile(TransformerNet_path): print("=> loading checkpoint '{}'".format(TransformerNet_path)) TransformerNet_par = torch.load(TransformerNet_path) for k in list(TransformerNet_par.keys()): if re.search(r'in\d+\.running_(mean|var)$', k): del TransformerNet_par[k] transformer.load_state_dict(TransformerNet_par) print("=> loaded checkpoint '{}'".format(TransformerNet_path)) else: print("=> no checkpoint found at '{}'".format(TransformerNet_path)) vgg = Vgg16(requires_grad=False).to(device) style = Image.open(style_image_path) style = transform(style) style = style.repeat(batch_size, 1, 1, 1).to(device) features_style = vgg(utils.normalize_batch(style)) gram_style = [utils.gram_matrix(y) for y in features_style] model_fcrn = FCRN_for_transfer(batch_size=batch_size, requires_grad=False).to(device) model_fcrn_par = torch.load(FCRN_path) #start_epoch = model_fcrn_par['epoch'] model_fcrn.load_state_dict(model_fcrn_par['state_dict']) print("=> loaded checkpoint '{}' (epoch {})".format( FCRN_path, model_fcrn_par['epoch'])) for e in range(epochs): transformer.train() agg_content_loss = 0. agg_depth_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) count += n_batch optimizer.zero_grad() x = x.to(device) y = transformer(x) y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y) features_x = vgg(x) depth_y = model_fcrn(y) depth_x = model_fcrn(x) content_loss = content_weight * mse_loss(features_y.relu2_2, features_x.relu2_2) depth_loss = depth_weight * mse_loss(depth_y, depth_x) style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = utils.gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[:n_batch, :, :]) style_loss *= style_weight total_loss = content_loss + depth_loss + style_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.item() agg_depth_loss += depth_loss.item() agg_style_loss += style_loss.item() if (batch_id + 1) % log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tdepth: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_depth_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1)) print(mesg) if checkpoint_model_dir is not None and ( batch_id + 1) % checkpoint_interval == 0: transformer.eval().cpu() ckpt_model_filename = "ckpt_epoch_" + str( e) + "_batch_id_" + str(batch_id + 1) + ".pth" ckpt_model_path = os.path.join(checkpoint_model_dir, ckpt_model_filename) torch.save(transformer.state_dict(), ckpt_model_path) transformer.to(device).train() # save model transformer.eval().cpu() save_model_filename = "epoch_" + str(epochs) + "_" + str( time.ctime()).replace(' ', '_') + "_" + str( content_weight) + "_" + str(style_weight) + ".model" save_model_path = os.path.join(save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(args): device = torch.device("cuda" if args.cuda else "cpu") np.random.seed(args.seed) torch.manual_seed(args.seed) transform = transforms.Compose([ transforms.Resize(args.image_size), transforms.CenterCrop(args.image_size), transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255)) ]) train_dataset = datasets.ImageFolder(args.dataset, transform) train_loader = DataLoader(train_dataset, batch_size=args.batch_size) transformer = TransformerNet().to(device) optimizer = Adam(transformer.parameters(), args.lr) mse_loss = torch.nn.MSELoss() vgg = Vgg16(requires_grad=False).to(device) style_transform = transforms.Compose( [transforms.ToTensor(), transforms.Lambda(lambda x: x.mul(255))]) style = utils.load_image(args.style_image, size=args.style_size) style = style_transform(style) style = style.repeat(args.batch_size, 1, 1, 1).to(device) features_style = vgg(utils.normalize_batch(style)) gram_style = [utils.gram_matrix(y) for y in features_style] for e in range(args.epochs): transformer.train() agg_content_loss = 0. agg_style_loss = 0. count = 0 for batch_id, (x, _) in enumerate(train_loader): n_batch = len(x) count += n_batch optimizer.zero_grad() x = x.to(device) y = transformer(x) y = utils.normalize_batch(y) x = utils.normalize_batch(x) features_y = vgg(y) features_x = vgg(x) content_loss = args.content_weight * mse_loss( features_y.relu2_2, features_x.relu2_2) style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gm_y = utils.gram_matrix(ft_y) style_loss += mse_loss(gm_y, gm_s[:n_batch, :, :]) style_loss *= args.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() agg_content_loss += content_loss.item() agg_style_loss += style_loss.item() if (batch_id + 1) % args.log_interval == 0: mesg = "{}\tEpoch {}:\t[{}/{}]\tcontent: {:.6f}\tstyle: {:.6f}\ttotal: {:.6f}".format( time.ctime(), e + 1, count, len(train_dataset), agg_content_loss / (batch_id + 1), agg_style_loss / (batch_id + 1), (agg_content_loss + agg_style_loss) / (batch_id + 1)) print(mesg) if args.checkpoint_model_dir is not None and ( batch_id + 1) % args.checkpoint_interval == 0: transformer.eval().cpu() ckpt_model_filename = "ckpt_epoch_" + str( e) + "_batch_id_" + str(batch_id + 1) + ".pth" ckpt_model_path = os.path.join(args.checkpoint_model_dir, ckpt_model_filename) torch.save(transformer.state_dict(), ckpt_model_path) transformer.to(device).train() # save model transformer.eval().cpu() save_model_filename = "epoch_" + str(args.epochs) + "_" + str( time.ctime()).replace(' ', '_') + "_" + str( args.content_weight) + "_" + str(args.style_weight) + ".model" save_model_path = os.path.join(args.save_model_dir, save_model_filename) torch.save(transformer.state_dict(), save_model_path) print("\nDone, trained model saved at", save_model_path)
def train(**kwargs): opt = Config() for k_, v_ in kwargs.items(): setattr(opt, k_, v_) device = t.device('cuda') if opt.use_gpu else t.device("cpu") vis = util.Visualizer(opt.env) transfroms = tv.transforms.Compose([ tv.transforms.Resize(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), tv.transforms.Lambda(lambda x: x * 255) ]) dataset = tv.datasets.ImageFolder(opt.data_root, transfroms) dataloader = data.DataLoader(dataset, opt.batch_size) transform = TransformerNet() if opt.model_path: transform.load_state_dict( t.load(opt.model_path, map_location=lambda _s, _: _s)) transform = transform.to(device) vgg = Vgg16().eval() vgg.to(device) for param in vgg.parameters(): param.requires_grad = False optimizer = t.optim.Adam(transform.parameters(), opt.lr) style = util.get_style_data(opt.style_path) vis.img("style", (style.data[0] * 0.225 + 0.45).clamp(min=0, max=1)) style = style.to(device) with t.no_grad(): features_style = vgg(style) gram_style = [util.gram_matrix(y) for y in features_style] style_meter = tnt.meter.AverageValueMeter() content_meter = tnt.meter.AverageValueMeter() for epoch in range(opt.epoches): content_meter.reset() style_meter.reset() for ii, (x, _) in tqdm.tqdm(enumerate(dataloader)): # 训练 optimizer.zero_grad() x = x.to(device) y = t.nn.parallel.data_parallel(transform, x, [0, 1]) y = util.normalize_batch(y) x = util.normalize_batch(x) features_y = vgg(y) features_x = vgg(x) content_loss = opt.content_weight * F.mse_loss( features_y.relu2_2, features_x.relu2_2) style_loss = 0. for ft_y, gm_s in zip(features_y, gram_style): gram_y = util.gram_matrix(ft_y) style_loss += F.mse_loss(gram_y, gm_s.expand_as(gram_y)) style_loss *= opt.style_weight total_loss = content_loss + style_loss total_loss.backward() optimizer.step() content_meter.add(content_loss.item()) style_meter.add(style_loss.item()) if (ii + 1) % opt.plot_every == 0: if os.path.exists(opt.debug_file): ipdb.set_trace() vis.plot("content_loss", content_meter.value()[0]) vis.plot("style_loss", style_meter.value()[0]) vis.img("output", (y.data.cpu()[0] * 0.255 + 0.45).clamp(min=0, max=1)) vis.img("input", (x.data.cpu()[0] * 0.255 + 0.45).clamp(min=0, max=1)) vis.save([opt.env]) t.save(transform.state_dict(), 'checkpoints/' + time.ctime() + '%s_style.pth' % epoch)