def train(): dataset = torchvision.datasets.ImageFolder(conf.data_path, transform=transforms) dataloader = torch.utils.data.DataLoader(dataset=dataset, batch_size=conf.batch_size, shuffle=True, drop_last=True) netG = NetG(conf.ngf, conf.nz) netD = NetD(conf.ndf) criterion = nn.BCELoss() optimizerG = torch.optim.Adam(netG.parameters(), lr=conf.lr, betas=(conf.beta1, 0.999)) optimizerD = torch.optim.Adam(netD.parameters(), lr=conf.lr, betas=(conf.beta1, 0.999)) label = torch.FloatTensor(conf.batch_size) real_label = 1 fake_label = 0 for epoch in range(1, conf.epoch + 1): for i, (imgs, _) in enumerate(dataloader): #step1:固定G,训练D optimizerD.zero_grad() output = netD(imgs) #让D尽可能把真图片识别为1 label.data.fill_(real_label) errD_real = criterion(output, label) errD_real.backward() #让D尽可能把假图判别为0 label.data.fill_(fake_label) noise = torch.randn(conf.batch_size, conf.nz, 1, 1) fake = netG(noise) #生成假图 output = netD(fake.detach()) #避免梯度传到G,因为G不用更新 errD_fake = criterion(output, label) errD_fake.backward() errD = errD_fake + errD_real optimizerD.step() #step2:固定判别器D,训练生成器G optimizerG.zero_grad() label.data.fill_(real_label) #让D尽可能把G生成的假图判别为1 output = netD(fake) errG = criterion(output, label) errG.backward() optimizerG.step() if i % 4 == 0: rate = i * 1.0 / len(dataloader) * 100 logger.info( "epoch={}, i={}, N={}, rate={}%, errD={}, errG={}".format( epoch, i, len(dataloader), rate, errD, errG)) #end-for save_image(fake.data, '%s/fake_samples_epoch_%03d.png' % (conf.checkpoints, epoch), normalize=True) torch.save(netG.state_dict(), '%s/netG_%03d.pth' % (conf.checkpoints, epoch)) torch.save(netD.state_dict(), '%s/netD_%03d.pth' % (conf.checkpoints, epoch))
class GANAgent(object): def __init__(self, input_size, output_size, num_env, num_step, gamma, lam=0.95, learning_rate=1e-4, ent_coef=0.01, clip_grad_norm=0.5, epoch=3, batch_size=128, ppo_eps=0.1, update_proportion=0.25, use_gae=True, use_cuda=False, use_noisy_net=False, hidden_dim=512): self.model = CnnActorCriticNetwork(input_size, output_size, use_noisy_net) self.num_env = num_env self.output_size = output_size self.input_size = input_size self.num_step = num_step self.gamma = gamma self.lam = lam self.epoch = epoch self.batch_size = batch_size self.use_gae = use_gae self.ent_coef = ent_coef self.ppo_eps = ppo_eps self.clip_grad_norm = clip_grad_norm self.update_proportion = update_proportion self.device = torch.device('cuda' if use_cuda else 'cpu') self.netG = NetG(z_dim=hidden_dim) #(input_size, z_dim=hidden_dim) self.netD = NetD(z_dim=1) self.netG.apply(weights_init) self.netD.apply(weights_init) self.optimizer_policy = optim.Adam(list(self.model.parameters()), lr=learning_rate) self.optimizer_G = optim.Adam(list(self.netG.parameters()), lr=learning_rate, betas=(0.5, 0.999)) self.optimizer_D = optim.Adam(list(self.netD.parameters()), lr=learning_rate, betas=(0.5, 0.999)) self.netG = self.netG.to(self.device) self.netD = self.netD.to(self.device) self.model = self.model.to(self.device) def reconstruct(self, state): state = torch.Tensor(state).to(self.device) state = state.float() reconstructed = self.vae(state.unsqueeze(0))[0].squeeze(0) return reconstructed.detach().cpu().numpy() def get_action(self, state): state = torch.Tensor(state).to(self.device) state = state.float() policy, value_ext, value_int = self.model(state) action_prob = F.softmax(policy, dim=-1).data.cpu().numpy() action = self.random_choice_prob_index(action_prob) return action, value_ext.data.cpu().numpy().squeeze( ), value_int.data.cpu().numpy().squeeze(), policy.detach() @staticmethod def random_choice_prob_index(p, axis=1): r = np.expand_dims(np.random.rand(p.shape[1 - axis]), axis=axis) return (p.cumsum(axis=axis) > r).argmax(axis=axis) def compute_intrinsic_reward(self, obs): obs = torch.FloatTensor(obs).to(self.device) #embedding = self.vae.representation(obs) #reconstructed_embedding = self.vae.representation(self.vae(obs)[0]) # why use index[0] reconstructed_img, embedding, reconstructed_embedding = self.netG(obs) intrinsic_reward = (embedding - reconstructed_embedding ).pow(2).sum(1) / 2 # Not use reconstructed loss return intrinsic_reward.detach().cpu().numpy() def train_model(self, s_batch, target_ext_batch, target_int_batch, y_batch, adv_batch, next_obs_batch, old_policy): s_batch = torch.FloatTensor(s_batch).to(self.device) target_ext_batch = torch.FloatTensor(target_ext_batch).to(self.device) target_int_batch = torch.FloatTensor(target_int_batch).to(self.device) y_batch = torch.LongTensor(y_batch).to(self.device) adv_batch = torch.FloatTensor(adv_batch).to(self.device) next_obs_batch = torch.FloatTensor(next_obs_batch).to(self.device) sample_range = np.arange(len(s_batch)) #reconstruction_loss = nn.MSELoss(reduction='none')] l_adv = nn.MSELoss(reduction='none') l_con = nn.L1Loss(reduction='none') l_enc = nn.MSELoss(reduction='none') l_bce = nn.BCELoss(reduction='none') with torch.no_grad(): policy_old_list = torch.stack(old_policy).permute( 1, 0, 2).contiguous().view(-1, self.output_size).to(self.device) m_old = Categorical(F.softmax(policy_old_list, dim=-1)) log_prob_old = m_old.log_prob(y_batch) # ------------------------------------------------------------ #recon_losses = np.array([]) #kld_losses = np.array([]) mean_err_g_adv_per_batch = np.array([]) mean_err_g_con_per_batch = np.array([]) mean_err_g_enc_per_batch = np.array([]) mean_err_d_per_batch = np.array([]) for i in range(self.epoch): np.random.shuffle(sample_range) for j in range(int(len(s_batch) / self.batch_size)): sample_idx = sample_range[self.batch_size * j:self.batch_size * (j + 1)] # -------------------------------------------------------------------------------- # for generative curiosity (GAN loss) #gen_next_state, mu, logvar = self.vae(next_obs_batch[sample_idx]) ############### netG forward ############################################## gen_next_state, latent_i, latent_o = self.netG( next_obs_batch[sample_idx]) ############### netD forward ############################################## pred_real, feature_real = self.netD(next_obs_batch[sample_idx]) pred_fake, feature_fake = self.netD(gen_next_state) #d = len(gen_next_state.shape) #recon_loss = reconstruction_loss(gen_next_state, next_obs_batch[sample_idx]).mean(axis=list(range(1, d))) ############### netG backward ############################################# self.optimizer_G.zero_grad() err_g_adv_per_img = l_adv( self.netD(next_obs_batch[sample_idx])[1], self.netD(gen_next_state)[1]).mean( axis=list(range(1, len(feature_real.shape)))) err_g_con_per_img = l_con( next_obs_batch[sample_idx], gen_next_state).mean( axis=list(range(1, len(gen_next_state.shape)))) err_g_enc_per_img = l_enc(latent_i, latent_o).mean(-1) #kld_loss = -0.5 * (1 + logvar - mu.pow(2) - logvar.exp()).sum(axis=1) # TODO: keep this proportion of experience used for VAE update? # Proportion of experience used for VAE update img_num = len(err_g_con_per_img) mask = torch.rand(img_num).to(self.device) mask = (mask < self.update_proportion).type( torch.FloatTensor).to(self.device) mean_err_g_adv = (err_g_adv_per_img * mask).sum() / torch.max( mask.sum(), torch.Tensor([1]).to(self.device)) mean_err_g_con = (err_g_con_per_img * mask).sum() / torch.max( mask.sum(), torch.Tensor([1]).to(self.device)) mean_err_g_enc = (err_g_enc_per_img * mask).sum() / torch.max( mask.sum(), torch.Tensor([1]).to(self.device)) # hyperparameter weights: w_adv = 1 w_con = 50 w_enc = 1 mean_err_g = mean_err_g_adv * w_adv +\ mean_err_g_con * w_con +\ mean_err_g_enc * w_enc mean_err_g.backward(retain_graph=True) self.optimizer_G.step() mean_err_g_adv_per_batch = np.append( mean_err_g_adv_per_batch, mean_err_g_adv.detach().cpu().numpy()) mean_err_g_con_per_batch = np.append( mean_err_g_con_per_batch, mean_err_g_con.detach().cpu().numpy()) mean_err_g_enc_per_batch = np.append( mean_err_g_enc_per_batch, mean_err_g_enc.detach().cpu().numpy()) ############## netD backward ############################################## self.optimizer_D.zero_grad() real_label = torch.ones_like(pred_real).to(self.device) fake_label = torch.zeros_like(pred_fake).to(self.device) err_d_real_per_img = l_bce(pred_real, real_label) err_d_fake_per_img = l_bce(pred_fake, fake_label) mean_err_d_real = (err_d_real_per_img * mask).sum() / torch.max( mask.sum(), torch.Tensor([1]).to(self.device)) mean_err_d_fake = (err_d_fake_per_img * mask).sum() / torch.max( mask.sum(), torch.Tensor([1]).to(self.device)) mean_err_d = (mean_err_d_real + mean_err_d_fake) / 2 mean_err_d.backward() self.optimizer_D.step() mean_err_d_per_batch = np.append( mean_err_d_per_batch, mean_err_d.detach().cpu().numpy()) if mean_err_d.item() < 1e-5: self.netD.apply(weights_init) print('Reloading net d') ############# policy update ############################################### policy, value_ext, value_int = self.model(s_batch[sample_idx]) m = Categorical(F.softmax(policy, dim=-1)) log_prob = m.log_prob(y_batch[sample_idx]) ratio = torch.exp(log_prob - log_prob_old[sample_idx]) surr1 = ratio * adv_batch[sample_idx] surr2 = torch.clamp(ratio, 1.0 - self.ppo_eps, 1.0 + self.ppo_eps) * adv_batch[sample_idx] actor_loss = -torch.min(surr1, surr2).mean() critic_ext_loss = F.mse_loss(value_ext.sum(1), target_ext_batch[sample_idx]) critic_int_loss = F.mse_loss(value_int.sum(1), target_int_batch[sample_idx]) critic_loss = critic_ext_loss + critic_int_loss entropy = m.entropy().mean() self.optimizer_policy.zero_grad() loss = actor_loss + 0.5 * critic_loss - self.ent_coef * entropy loss.backward() #global_grad_norm_(list(self.model.parameters())+list(self.vae.parameters())) do we need this step #global_grad_norm_(list(self.model.parameter())) or just norm policy self.optimizer_poilicy.step() return mean_err_g_adv_per_batch, mean_err_g_con_per_batch, mean_err_g_enc_per_batch, mean_err_d_per_batch def train_just_vae(self, s_batch, next_obs_batch): s_batch = torch.FloatTensor(s_batch).to(self.device) next_obs_batch = torch.FloatTensor(next_obs_batch).to(self.device) sample_range = np.arange(len(s_batch)) l_adv = nn.MSELoss(reduction='none') l_con = nn.L1Loss(reduction='none') l_enc = nn.MSELoss(reduction='none') l_bce = nn.BCELoss(reduction='none') mean_err_g_adv_per_batch = np.array([]) mean_err_g_con_per_batch = np.array([]) mean_err_g_enc_per_batch = np.array([]) mean_err_d_per_batch = np.array([]) for i in range(self.epoch): np.random.shuffle(sample_range) for j in range(int(len(s_batch) / self.batch_size)): sample_idx = sample_range[self.batch_size * j:self.batch_size * (j + 1)] ############### netG forward ############################################## gen_next_state, latent_i, latent_o = self.netG( next_obs_batch[sample_idx]) ############### netD forward ############################################## pred_real, feature_real = self.netD(next_obs_batch[sample_idx]) pred_fake, feature_fake = self.netD(gen_next_state) #d = len(gen_next_state.shape) #recon_loss = reconstruction_loss(gen_next_state, next_obs_batch[sample_idx]).mean(axis=list(range(1, d))) ############### netG backward ############################################# self.optimizer_G.zero_grad() err_g_adv_per_img = l_adv( self.netD(next_obs_batch[sample_idx])[1], self.netD(gen_next_state)[1]).mean( axis=list(range(1, len(feature_real.shape)))) err_g_con_per_img = l_con( next_obs_batch[sample_idx], gen_next_state).mean( axis=list(range(1, len(gen_next_state.shape)))) err_g_enc_per_img = l_enc(latent_i, latent_o).mean(-1) #kld_loss = -0.5 * (1 + logvar - mu.pow(2) - logvar.exp()).sum(axis=1) # TODO: keep this proportion of experience used for VAE update? # Proportion of experience used for VAE update img_num = len(err_g_con_per_img) mask = torch.rand(img_num).to(self.device) mask = (mask < self.update_proportion).type( torch.FloatTensor).to(self.device) mean_err_g_adv = (err_g_adv_per_img * mask).sum() / torch.max( mask.sum(), torch.Tensor([1]).to(self.device)) mean_err_g_con = (err_g_con_per_img * mask).sum() / torch.max( mask.sum(), torch.Tensor([1]).to(self.device)) mean_err_g_enc = (err_g_enc_per_img * mask).sum() / torch.max( mask.sum(), torch.Tensor([1]).to(self.device)) # hyperparameter weights: w_adv = 1 w_con = 50 w_enc = 1 mean_err_g = mean_err_g_adv * w_adv +\ mean_err_g_con * w_con +\ mean_err_g_enc * w_enc mean_err_g.backward(retain_graph=True) self.optimizer_G.step() mean_err_g_adv_per_batch = np.append( mean_err_g_adv_per_batch, mean_err_g_adv.detach().cpu().numpy()) mean_err_g_con_per_batch = np.append( mean_err_g_con_per_batch, mean_err_g_con.detach().cpu().numpy()) mean_err_g_enc_per_batch = np.append( mean_err_g_enc_per_batch, mean_err_g_enc.detach().cpu().numpy()) ############## netD backward ############################################## self.optimizer_D.zero_grad() real_label = torch.ones_like(pred_real).to(self.device) fake_label = torch.zeros_like(pred_fake).to(self.device) err_d_real_per_img = l_bce(pred_real, real_label) err_d_fake_per_img = l_bce(pred_fake, fake_label) mean_err_d_real = (err_d_real_per_img * mask).sum() / torch.max( mask.sum(), torch.Tensor([1]).to(self.device)) mean_err_d_fake = (err_d_fake_per_img * mask).sum() / torch.max( mask.sum(), torch.Tensor([1]).to(self.device)) mean_err_d = (mean_err_d_real + mean_err_d_fake) / 2 mean_err_d.backward() self.optimizer_D.step() mean_err_d_per_batch = np.append( mean_err_d_per_batch, mean_err_d.detach().cpu().numpy()) return mean_err_g_adv_per_batch, mean_err_g_con_per_batch, mean_err_g_enc_per_batch, mean_err_d_per_batch
def train(**kwargs): for k_, v_ in kwargs.items(): setattr(opt, k_, v_) device = t.device("cuda") if opt.gpu else t.device("cpu") # 数据处理,输出规范为-1~1 transforms = tv.transforms.Compose([ tv.transforms.Resize(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), tv.transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)) ]) dataset = tv.datasets.ImageFolder(opt.data_path, transform=transforms) dataloader = t.utils.data.DataLoader(dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_workers, drop_last=True) # 网络 netg, netd = NetG(opt), NetD(opt) map_location = lambda storage, loc: storage if opt.netd_path: netd.load_state_dict(t.load(opt.netd_path, map_location=map_location)) if opt.netg_path: netg.load_state_dict(t.load(opt.netg_path, map_location=map_location)) netd.to(device) netg.to(device) # 定义优化器和损失 optimizer_g = t.optim.Adam(netg.parameters(), opt.lr1, betas=(opt.beta1, 0.999)) optimizer_d = t.optim.Adam(netd.parameters(), opt.lr2, betas=(opt.beta1, 0.999)) criterion = t.nn.BCELoss() # 真图片label为1,假图片label为0, noise为生成网络的输入 true_labels = t.ones(opt.batch_size).to(device) fake_labels = t.zeros(opt.batch_size).to(device) fix_noises = t.randn(opt.batch_size, opt.nz, 1, 1).to(device) noises = t.randn(opt.batch_size, opt.nz, 1, 1).to(device) # 用来结果的均值和标准差 errord_meter = AverageValueMeter() errorg_meter = AverageValueMeter() epochs = range(opt.max_epoch) for epoch in iter(epochs): for ii, (img, _) in tqdm.tqdm(enumerate(dataloader)): real_img = img.to(device) if ii % opt.d_every == 0: # 训练判别器 optimizer_d.zero_grad() ## 尽可能把真图片判别为正 output = netd(real_img) error_d_real = criterion(output, true_labels) error_d_real.backward() ## 尽可能把假图片判断为错误 noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) # 使用detach来关闭G求梯度,加速训练 fake_img = netg(noises).detach() output = netd(fake_img) error_d_fake = criterion(output, fake_labels) error_d_fake.backward() optimizer_d.step() error_d = error_d_fake + error_d_real errord_meter.add(error_d.item()) if ii % opt.g_every == 0: # 训练生成器 optimizer_g.zero_grad() noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises) output = netd(fake_img) # 尽可能把假的图片也判别为1 error_g = criterion(output, true_labels) error_g.backward() optimizer_g.step() errorg_meter.add(error_g.item()) # 可视化 # 保存模型、图片 if (epoch + 1) % opt.save_every == 0: fix_fake_imgs = netg(fix_noises) tv.utils.save_image(fix_fake_imgs.data[:64], "%s%s.png" % (opt.save_path, epoch), normalize=True, range=(-1, 1)) t.save(netd.state_dict(), r"./checkpoints/netd_%s.pth" % epoch) t.save(netg.state_dict(), r"./checkpoints/netg_%s.pth" % epoch) errord_meter.reset() errorg_meter.reset()
def train(**kwargs): for k_, v_ in kwargs.items(): setattr(opt, k_, v_) device = t.device('cuda') if opt.gpu else t.device('cpu') # 可视化 if opt.vis: from visualize import Visualizer vis = Visualizer(opt.env) # 数据 transforms = tv.transforms.Compose([ tv.transforms.Resize(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), tv.transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)) ]) dataset = tv.datasets.ImageFolder(opt.data_path, transforms) dataloader = t.utils.data.DataLoader(dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_workers, drop_last=True) netg, netd = NetG(opt), NetD(opt) if opt.netd_path: netd.load_state_dict( t.load(opt.netd_path, map_location=t.device('cpu'))) if opt.netg_path: netg.load_state_dict( t.load(opt.netg_path, map_location=t.device('cpu'))) netd.to(device) netg.to(device) # 定义优化器和损失 optimizer_g = t.optim.Adam(netg.parameters(), opt.lr1, betas=(opt.beta1, 0.999)) optimizer_d = t.optim.Adam(netd.parameters(), opt.lr1, betas=(opt.beta1, 0.999)) criterion = t.nn.BCELoss().to(device) # 真图片label为1,假图片label为0 # noise为生成网络的输入 true_labels = t.ones(opt.batch_size).to(device) fake_labels = t.zeros(opt.batch_size).to(device) fix_noises = t.randn(opt.batch_size, opt.nz, 1, 1).to(device) noises = t.randn(opt.batch_size, opt.nz, 1, 1).to(device) errord = 0 errorg = 0 epochs = range(opt.max_epoch) for epoch in epochs: for ii, (img, _) in tqdm(enumerate(dataloader)): real_img = img.to(device) if (ii + 1) % opt.d_every == 0: # 训练判别器 optimizer_d.zero_grad() # 把真图片判断为正确 output = netd(real_img) error_d_real = criterion(output, true_labels) error_d_real.backward() # 把假图片(netg通过噪声生成的图片)判断为错误 noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises).detach() output = netd(fake_img) error_d_fake = criterion(output, fake_labels) error_d_fake.backward() optimizer_d.step() errord += (error_d_fake + error_d_real).item() if (ii + 1) % opt.g_every == 0: # 训练生成器 optimizer_g.zero_grad() noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises) output = netd(fake_img) error_g = criterion(output, true_labels) error_g.backward() optimizer_g.step() errorg += error_g.item() if opt.vis and (ii + 1) % opt.plot_every == 0: fix_fake_imgs = netg(fix_noises) vis.images(fix_fake_imgs.detach().cpu().numpy()[:64] * 0.5 + 0.5, win='fixfake') vis.images(real_img.data.cpu().numpy()[:64] * 0.5 + 0.5, win='real') vis.plot('errord', errord / (opt.plot_every)) vis.plot('errorg', errorg / (opt.plot_every)) errord = 0 errorg = 0 if (epoch + 1) % opt.save_every == 0: tv.utils.save_image(fix_fake_imgs.data[:64], '%s%s.png' % (opt.save_path, epoch), normalize=True, range=(-1, 1)) t.save(netd.state_dict(), 'checkpoints/netd_%s.pth' % (epoch + 1)) t.save(netg.state_dict(), 'checkpoints/netg_%s.pth' % (epoch + 1))
def train(**kwargs): for k_, v_ in kwargs.items(): setattr(opt, k_, v_) device=t.device('cuda') if opt.gpu else t.device('cpu') if opt.vis: from visualize import Visualizer vis = Visualizer(opt.env) # 数据 transforms = tv.transforms.Compose([ tv.transforms.Resize(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), tv.transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)) ]) dataset = tv.datasets.ImageFolder(opt.data_path, transform=transforms) dataloader = t.utils.data.DataLoader(dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_workers, drop_last=True ) # 网络 netg, netd = NetG(opt), NetD(opt) map_location = lambda storage, loc: storage if opt.netd_path: netd.load_state_dict(t.load(opt.netd_path, map_location=map_location)) if opt.netg_path: netg.load_state_dict(t.load(opt.netg_path, map_location=map_location)) netd.to(device) netg.to(device) # 定义优化器和损失 optimizer_g = t.optim.Adam(netg.parameters(), opt.lr1, betas=(opt.beta1, 0.999)) optimizer_d = t.optim.Adam(netd.parameters(), opt.lr2, betas=(opt.beta1, 0.999)) criterion = t.nn.BCELoss().to(device) # 真图片label为1,假图片label为0 # noises为生成网络的输入 true_labels = t.ones(opt.batch_size).to(device) fake_labels = t.zeros(opt.batch_size).to(device) fix_noises = t.randn(opt.batch_size, opt.nz, 1, 1).to(device) noises = t.randn(opt.batch_size, opt.nz, 1, 1).to(device) errord_meter = AverageValueMeter() errorg_meter = AverageValueMeter() epochs = range(opt.max_epoch) for epoch in iter(epochs): for ii, (img, _) in tqdm.tqdm(enumerate(dataloader)): real_img = img.to(device) if ii % opt.d_every == 0: # 训练判别器 optimizer_d.zero_grad() ## 尽可能的把真图片判别为正确 output = netd(real_img) error_d_real = criterion(output, true_labels) error_d_real.backward() ## 尽可能把假图片判别为错误 noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises).detach() # 根据噪声生成假图 output = netd(fake_img) error_d_fake = criterion(output, fake_labels) error_d_fake.backward() optimizer_d.step() error_d = error_d_fake + error_d_real errord_meter.add(error_d.item()) if ii % opt.g_every == 0: # 训练生成器 optimizer_g.zero_grad() noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises) output = netd(fake_img) error_g = criterion(output, true_labels) error_g.backward() optimizer_g.step() errorg_meter.add(error_g.item()) if opt.vis and ii % opt.plot_every == opt.plot_every - 1: ## 可视化 if os.path.exists(opt.debug_file): ipdb.set_trace() fix_fake_imgs = netg(fix_noises) vis.images(fix_fake_imgs.detach().cpu().numpy()[:64] * 0.5 + 0.5, win='fixfake') vis.images(real_img.data.cpu().numpy()[:64] * 0.5 + 0.5, win='real') vis.plot('errord', errord_meter.value()[0]) vis.plot('errorg', errorg_meter.value()[0]) if (epoch+1) % opt.save_every == 0: # 保存模型、图片 tv.utils.save_image(fix_fake_imgs.data[:64], '%s/%s.png' % (opt.save_path, epoch), normalize=True, range=(-1, 1)) t.save(netd.state_dict(), 'checkpoints/netd_%s.pth' % epoch) t.save(netg.state_dict(), 'checkpoints/netg_%s.pth' % epoch) errord_meter.reset() errorg_meter.reset()
def train(**kwargs): for k_, v_ in kwargs.items(): setattr(opt, k_, v_) device = t.device('cuda') if opt.gpu else t.device('cpu') # if opt.vis: # from visualize import Visualizer # vis = Visualizer(opt.env) # 数据 transforms = tv.transforms.Compose([ tv.transforms.Resize(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), tv.transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)) ]) dataset = tv.datasets.ImageFolder(root=opt.data_path, transform=transforms) dataloader = t.utils.data.DataLoader(dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_workers, drop_last=True) # 网络 netg, netd = NetG(opt), NetD(opt) map_location = lambda storage, loc: storage if opt.netd_path: netd.load_state_dict(t.load(opt.netd_path, map_location=map_location)) if opt.netg_path: netg.load_state_dict(t.load(opt.netg_path, map_location=map_location)) netd.to(device) netg.to(device) # 定义优化器和损失 optimizer_g = t.optim.Adam(netg.parameters(), opt.lr1, betas=(opt.beta1, 0.999)) optimizer_d = t.optim.Adam(netd.parameters(), opt.lr2, betas=(opt.beta1, 0.999)) criterion = t.nn.BCELoss().to(device) # 真图片label为1,假图片label为0 # noises为生成网络的输入 true_labels = t.ones(opt.batch_size).to(device) fake_labels = t.zeros(opt.batch_size).to(device) fix_noises = t.randn(opt.batch_size, opt.nz, 1, 1).to(device) noises = t.randn(opt.batch_size, opt.nz, 1, 1).to(device) errord_meter = AverageValueMeter() errorg_meter = AverageValueMeter() epochs = range(opt.max_epoch) for epoch in iter(epochs): for ii, (img, _) in tqdm.tqdm(enumerate(dataloader)): real_img = img.to(device) if ii % opt.d_every == 0: # 训练判别器 optimizer_d.zero_grad() ## 尽可能的把真图片判别为正确 output = netd(real_img) error_d_real = criterion(output, true_labels) error_d_real.backward() ## 尽可能把假图片判别为错误 noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises).detach() # 根据噪声生成假图 output = netd(fake_img) error_d_fake = criterion(output, fake_labels) error_d_fake.backward() optimizer_d.step() error_d = error_d_fake + error_d_real errord_meter.add(error_d.item()) if ii % opt.g_every == 0: # 训练生成器 optimizer_g.zero_grad() noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises) output = netd(fake_img) error_g = criterion(output, true_labels) error_g.backward() optimizer_g.step() errorg_meter.add(error_g.item()) if opt.vis and ii % opt.plot_every == opt.plot_every - 1: ## 可视化 if os.path.exists(opt.debug_file): ipdb.set_trace() fix_fake_imgs = netg(fix_noises) vis.images(fix_fake_imgs.detach().cpu().numpy()[:64] * 0.5 + 0.5, win='fixfake') vis.images(real_img.data.cpu().numpy()[:64] * 0.5 + 0.5, win='real') vis.plot('errord', errord_meter.value()[0]) vis.plot('errorg', errorg_meter.value()[0]) if (epoch + 1) % opt.save_every == 0: # 保存模型、图片 tv.utils.save_image(fix_fake_imgs.data[:64], '%s/%s.png' % (opt.save_path, epoch), normalize=True, range=(-1, 1)) t.save(netd.state_dict(), 'checkpoints/netd_%s.pth' % epoch) t.save(netg.state_dict(), 'checkpoints/netg_%s.pth' % epoch) errord_meter.reset() errorg_meter.reset()
def train_network(): init_epoch = 0 best_f1 = 0 total_steps = 0 train_dir = ct.TRAIN_TXT val_dir = ct.VAL_TXT device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') torch.backends.cudnn.benchmark = True train_data = OSCD_TRAIN(train_dir) train_dataloader = DataLoader(train_data, batch_size=ct.BATCH_SIZE, shuffle=True) val_data = OSCD_TEST(val_dir) val_dataloader = DataLoader(val_data, batch_size=1, shuffle=False) netg = NetG(ct.ISIZE, ct.NC * 2, ct.NZ, ct.NDF, ct.EXTRALAYERS).to(device=device) netd = NetD(ct.ISIZE, ct.GT_C, 1, ct.NGF, ct.EXTRALAYERS).to(device=device) netg.apply(weights_init) netd.apply(weights_init) if ct.RESUME: assert os.path.exists(os.path.join(ct.WEIGHTS_SAVE_DIR, 'current_netG.pth')) \ and os.path.exists(os.path.join(ct.WEIGHTS_SAVE_DIR, 'current_netG.pth')), \ 'There is not found any saved weights' print("\nLoading pre-trained networks.") init_epoch = torch.load( os.path.join(ct.WEIGHTS_SAVE_DIR, 'current_netG.pth'))['epoch'] netg.load_state_dict( torch.load(os.path.join(ct.WEIGHTS_SAVE_DIR, 'current_netG.pth'))['model_state_dict']) netd.load_state_dict( torch.load(os.path.join(ct.WEIGHTS_SAVE_DIR, 'current_netD.pth'))['model_state_dict']) with open(os.path.join(ct.OUTPUTS_DIR, 'f1_score.txt')) as f: lines = f.readlines() best_f1 = float(lines[-2].strip().split(':')[-1]) print("\tDone.\n") l_adv = l2_loss l_con = nn.L1Loss() l_enc = l2_loss l_bce = nn.BCELoss() l_cos = cos_loss dice = DiceLoss() optimizer_d = optim.Adam(netd.parameters(), lr=ct.LR, betas=(0.5, 0.999)) optimizer_g = optim.Adam(netg.parameters(), lr=ct.LR, betas=(0.5, 0.999)) start_time = time.time() for epoch in range(init_epoch + 1, ct.EPOCH): loss_g = [] loss_d = [] netg.train() netd.train() epoch_iter = 0 for i, data in enumerate(train_dataloader): INPUT_SIZE = [ct.ISIZE, ct.ISIZE] x1, x2, gt = data x1 = x1.to(device, dtype=torch.float) x2 = x2.to(device, dtype=torch.float) gt = gt.to(device, dtype=torch.float) gt = gt[:, 0, :, :].unsqueeze(1) x = torch.cat((x1, x2), 1) epoch_iter += ct.BATCH_SIZE total_steps += ct.BATCH_SIZE real_label = torch.ones(size=(x1.shape[0], ), dtype=torch.float32, device=device) fake_label = torch.zeros(size=(x1.shape[0], ), dtype=torch.float32, device=device) #forward fake = netg(x) pred_real = netd(gt) pred_fake = netd(fake).detach() err_d_fake = l_bce(pred_fake, fake_label) err_g = l_con(fake, gt) err_g_total = ct.G_WEIGHT * err_g + ct.D_WEIGHT * err_d_fake pred_fake_ = netd(fake.detach()) err_d_real = l_bce(pred_real, real_label) err_d_fake_ = l_bce(pred_fake_, fake_label) err_d_total = (err_d_real + err_d_fake_) * 0.5 #backward optimizer_g.zero_grad() err_g_total.backward(retain_graph=True) optimizer_g.step() optimizer_d.zero_grad() err_d_total.backward() optimizer_d.step() errors = utils.get_errors(err_d_total, err_g_total) loss_g.append(err_g_total.item()) loss_d.append(err_d_total.item()) counter_ratio = float(epoch_iter) / len(train_dataloader.dataset) if (i % ct.DISPOLAY_STEP == 0 and i > 0): print( 'epoch:', epoch, 'iteration:', i, ' G|D loss is {}|{}'.format(np.mean(loss_g[-51:]), np.mean(loss_d[-51:]))) if ct.DISPLAY: utils.plot_current_errors(epoch, counter_ratio, errors, vis) utils.display_current_images(gt.data, fake.data, vis) utils.save_current_images(epoch, gt.data, fake.data, ct.IM_SAVE_DIR, 'training_output_images') with open(os.path.join(ct.OUTPUTS_DIR, 'train_loss.txt'), 'a') as f: f.write( 'after %s epoch, loss is %g,loss1 is %g,loss2 is %g,loss3 is %g' % (epoch, np.mean(loss_g), np.mean(loss_d), np.mean(loss_g), np.mean(loss_d))) f.write('\n') if not os.path.exists(ct.WEIGHTS_SAVE_DIR): os.makedirs(ct.WEIGHTS_SAVE_DIR) utils.save_weights(epoch, netg, optimizer_g, ct.WEIGHTS_SAVE_DIR, 'netG') utils.save_weights(epoch, netd, optimizer_d, ct.WEIGHTS_SAVE_DIR, 'netD') duration = time.time() - start_time print('training duration is %g' % duration) #val phase print('Validating.................') pretrained_dict = torch.load( os.path.join(ct.WEIGHTS_SAVE_DIR, 'current_netG.pth'))['model_state_dict'] device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') net = NetG(ct.ISIZE, ct.NC * 2, ct.NZ, ct.NDF, ct.EXTRALAYERS).to(device=device) net.load_state_dict(pretrained_dict, False) with net.eval() and torch.no_grad(): TP = 0 FN = 0 FP = 0 TN = 0 for k, data in enumerate(val_dataloader): x1, x2, label = data x1 = x1.to(device, dtype=torch.float) x2 = x2.to(device, dtype=torch.float) label = label.to(device, dtype=torch.float) label = label[:, 0, :, :].unsqueeze(1) x = torch.cat((x1, x2), 1) time_i = time.time() v_fake = net(x) tp, fp, tn, fn = eva.f1(v_fake, label) TP += tp FN += fn TN += tn FP += fp precision = TP / (TP + FP + 1e-8) oa = (TP + TN) / (TP + FN + TN + FP + 1e-8) recall = TP / (TP + FN + 1e-8) f1 = 2 * precision * recall / (precision + recall + 1e-8) if not os.path.exists(ct.BEST_WEIGHT_SAVE_DIR): os.makedirs(ct.BEST_WEIGHT_SAVE_DIR) if f1 > best_f1: best_f1 = f1 shutil.copy( os.path.join(ct.WEIGHTS_SAVE_DIR, 'current_netG.pth'), os.path.join(ct.BEST_WEIGHT_SAVE_DIR, 'netG.pth')) print('current F1: {}'.format(f1)) print('best f1: {}'.format(best_f1)) with open(os.path.join(ct.OUTPUTS_DIR, 'f1_score.txt'), 'a') as f: f.write('current epoch:{},current f1:{},best f1:{}'.format( epoch, f1, best_f1)) f.write('\n')
one = torch.FloatTensor([1]) mone = one * -1 realData = Variable(realData) z = Variable(z) netG = netG.cuda() netD = netD.cuda() z = z.cuda() realData = realData.cuda() one = one.cuda() mone = mone.cuda() # setup optimizer optimizerD = optim.RMSprop(netD.parameters(), lr=0.00005) # 0.00005 optimizerG = optim.RMSprop(netG.parameters(), lr=0.00005) if opt.ep != -1: netD.load_state_dict( torch.load(checkRoot + '/netD_epoch_' + str(opt.ep) + '.pth')) netG.load_state_dict( torch.load(checkRoot + '/netG_epoch_' + str(opt.ep) + '.pth')) # train ig = 0 for it in np.arange(iterNum) + opt.ep + 1: dataIter = iter(dataLoader) ib = 0 while ib < len(dataLoader): ############################ # (1) Update D network ########################### # train the discriminator Diters times
def train(): # change opt # for k_, v_ in kwargs.items(): # setattr(opt, k_, v_) device = torch.device('cuda') if torch.cuda.is_available else torch.device( 'cpu') if opt.vis: from visualizer import Visualizer vis = Visualizer(opt.env) # rescale to -1~1 transform = transforms.Compose([ transforms.Resize(opt.image_size), transforms.CenterCrop(opt.image_size), transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)) ]) dataset = datasets.ImageFolder(opt.data_path, transform=transform) dataloader = DataLoader(dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_workers, drop_last=True) netd = NetD(opt) netg = NetG(opt) map_location = lambda storage, loc: storage if opt.netd_path: netd.load_state_dict(torch.load(opt.netd_path), map_location=map_location) if opt.netg_path: netg.load_state_dict(torch.load(opt.netg_path), map_location=map_location) if torch.cuda.is_available(): netd.to(device) netg.to(device) # 定义优化器和损失 optimizer_g = torch.optim.Adam(netg.parameters(), opt.lr1, betas=(opt.beta1, 0.999)) optimizer_d = torch.optim.Adam(netd.parameters(), opt.lr2, betas=(opt.beta1, 0.999)) criterion = torch.nn.BCELoss().to(device) # 真label为1, noises是输入噪声 true_labels = Variable(torch.ones(opt.batch_size)) fake_labels = Variable(torch.zeros(opt.batch_size)) fix_noises = Variable(torch.randn(opt.batch_size, opt.nz, 1, 1)) noises = Variable(torch.randn(opt.batch_size, opt.nz, 1, 1)) errord_meter = AverageValueMeter() errorg_meter = AverageValueMeter() if torch.cuda.is_available(): netd.cuda() netg.cuda() criterion.cuda() true_labels, fake_labels = true_labels.cuda(), fake_labels.cuda() fix_noises, noises = fix_noises.cuda(), noises.cuda() for epoch in range(opt.max_epoch): print("epoch:", epoch, end='\r') # sys.stdout.flush() for ii, (img, _) in enumerate(dataloader): real_img = Variable(img) if torch.cuda.is_available(): real_img = real_img.cuda() # 训练判别器, real -> 1, fake -> 0 if (ii + 1) % opt.d_every == 0: # real optimizer_d.zero_grad() output = netd(real_img) # print(output.shape, true_labels.shape) error_d_real = criterion(output, true_labels) error_d_real.backward() # fake noises.data.copy_(torch.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises).detach() # 随机噪声生成假图 fake_output = netd(fake_img) error_d_fake = criterion(fake_output, fake_labels) error_d_fake.backward() # update optimizer optimizer_d.step() error_d = error_d_fake + error_d_real errord_meter.add(error_d.item()) # 训练生成器, 让生成器得到的图片能够被判别器判别为真 if (ii + 1) % opt.g_every == 0: optimizer_g.zero_grad() noises.data.copy_(torch.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises) fake_output = netd(fake_img) error_g = criterion(fake_output, true_labels) error_g.backward() optimizer_g.step() errorg_meter.add(error_g.item()) if opt.vis and ii % opt.plot_every == opt.plot_every - 1: # 进行可视化 # if os.path.exists(opt.debug_file): # import ipdb # ipdb.set_trace() fix_fake_img = netg(fix_noises) vis.images( fix_fake_img.detach().cpu().numpy()[:opt.batch_size] * 0.5 + 0.5, win='fixfake') vis.images(real_img.data.cpu().numpy()[:opt.batch_size] * 0.5 + 0.5, win='real') vis.plot('errord', errord_meter.value()[0]) vis.plot('errorg', errorg_meter.value()[0]) if (epoch + 1) % opt.save_every == 0: # 保存模型、图片 tv.utils.save_image(fix_fake_img.data[:opt.batch_size], '%s/%s.png' % (opt.save_path, epoch), normalize=True, range=(-1, 1)) torch.save(netd.state_dict(), 'checkpoints/netd_%s.pth' % epoch) torch.save(netg.state_dict(), 'checkpoints/netg_%s.pth' % epoch) errord_meter.reset() errorg_meter.reset()
# 打印网络模型 Print the model print(netG) print(netD) criterion = nn.BCELoss() # 初始化损失函数 Initialize BCELoss function # Create batch of latent vectors that we will use to visualize the progression of the generator fixed_noise = torch.randn(100, nz, 1, 1, device=device) # 评估真假的标签 真为1 假为0 Establish convention for real and fake labels during training real_label = 1 fake_label = 0 # Adam优化 Setup Adam optimizers for both G and D optimizerD = optim.Adam(netD.parameters(), lr=lr, betas=(beta1, 0.999)) optimizerG = optim.Adam(netG.parameters(), lr=lr, betas=(beta1, 0.999)) # 记录训练过程 Lists to keep track of progress G_losses = [] D_losses = [] iters = 0 print("开始训练...") # 对每一个epoch For each epoch for epoch in range(num_epochs): # 对于每一个batch For each batch in the dataloader for i, data in enumerate(dataloader, 0): ############################ # (1) Update D network: maximize log(D(x)) + log(1 - D(G(z))) ########################### # 使用真的图片训练鉴别器D 尽可能判别为1 Train with real batch
drop_last=True) netg = NetG(opt) netd = NetD(opt) # 加载已有的网络参数 if opt.netd_path: print('Loading netd...', end='') netd.load_state_dict(torch.load(opt.netd_path)) print('Successful!') if opt.netg_path: print('Loading netg...', end='') netg.load_state_dict(torch.load(opt.netg_path)) print('Successful!') optimizer_g = torch.optim.Adam(netg.parameters(), opt.lr_netg, betas=(opt.beta1, 0.999)) optimizer_d = torch.optim.Adam(netd.parameters(), opt.lr_netd, betas=(opt.beta1, 0.999)) criterion = torch.nn.BCELoss() # 真图片 label 为 1,假图片为 0 true_labels = Variable(torch.ones(opt.batch_size)) fake_labels = Variable(torch.zeros(opt.batch_size)) #fix_noises = Variable(torch.randn(opt.batch_size, opt.nz, 1, 1)) noises = Variable(torch.randn(opt.batch_size, opt.nz, 1, 1))
def train(**kwargs): for k_, v_ in kwargs.items(): setattr(opt, k_, v_) if opt.vis: from visualizer import Visualizer vis = Visualizer(opt.env) transforms = tv.transforms.Compose([ tv.transforms.Scale(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), tv.transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)) ]) dataset = tv.datasets.ImageFolder(opt.data_path, transform=transforms) dataloader = t.utils.data.DataLoader(dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_workers, drop_last=True) netg, netd = NetG(opt), NetD(opt) map_location = lambda storage, loc: storage if opt.netd_path: netd.load_state_dict(t.load(opt.netd_path, map_location=map_location)) if opt.netg_path: netg.load_state_dict(t.load(opt.netg_path, map_location=map_location)) optimizer_g = t.optim.Adam(netg.parameters(), opt.lr1, betas=(opt.beta1, 0.999)) optimizer_d = t.optim.Adam(netd.parameters(), opt.lr2, betas=(opt.beta1, 0.999)) criterion = t.nn.BCELoss() true_labels = Variable(t.ones(opt.batch_size)) fake_labels = Variable(t.zeros(opt.batch_size)) fix_noises = Variable(t.randn(opt.batch_size, opt.nz, 1, 1)) noises = Variable(t.randn(opt.batch_size, opt.nz, 1, 1)) errord_meter = AverageValueMeter() errorg_meter = AverageValueMeter() if opt.gpu: netd.cuda() netg.cuda() criterion.cuda() true_labels, fake_labels = true_labels.cuda(), fake_labels.cuda() fix_noises, noises = fix_noises.cuda(), noises.cuda() epochs = range(opt.max_epoch) for epoch in iter(epochs): for ii, (img, _) in tqdm.tqdm(enumerate(dataloader)): real_img = Variable(img) if opt.gpu: real_img = real_img.cuda() if ii % opt.d_every == 0: optimizer_d.zero_grad() output = netd(real_img) error_d_real = criterion(output, true_labels) error_d_real.backward() noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises).detach() output = netd(fake_img) error_d_fake = criterion(output, fake_labels) error_d_fake.backward() optimizer_d.step() error_d = error_d_fake + error_d_real error_meter.add(error_d.data[0]) if ii % opt.g_every == 0: optimizer_g.zero_grad() noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises) output = netd(fake_img).detach() error_g = criterion(output, true_labels) error_g.backward() optimizer_g.step() errorg_meter.add(error_g.data[0]) if opt.vis and ii % opt.plot_every == opt.plot_every - 1: if os.path.exists(opt.debug_file): ipdb.set_trace() fix_fake_imgs = netg(fix_noises) vis.images(fix_fake_imgs.data.cpu().numpy()[:64] * 0.5 + 0.5, win='fixfake') vis.images(real_img.data.cpu().numpy()[:64] * 0.5 + 0.5, win='real') vis.plot('errord', errord_meter.value()[0]) vis.plot('errorg', errorg_meter.value()[0]) if epoch % opt.decay_every == 0: tv.utils.save_image(fix_fake_imgs.data[:64], '%s/%s.png' % (opt.save_path, epoch), Normalize=True, range=(-1, 1)) t.save(netd.state_dict(), 'checkpoints/netd_%s.pth' % epoch) t.save(netg.state_dict(), 'checkpoints/netg_%s.pth' % epoch) errord_meter.reset() errorg_meter.reset()
def train(**kwargs): ''' 训练函数 :param kwargs: fire传进来的训练参数 :return: ''' opt.parse(kwargs) for k_,v_ in kwargs.items(): setattr(opt,k_,v_) if opt.vis: vis = Visualizer(opt.env) #step1:数据预处理 transforms = tv.transforms.Compose([ tv.transforms.Resize(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), tv.transforms.Normalize((0.5,0.5,0.5),(0.5,0.5,0.5)) ]) dataset = tv.datasets.ImageFolder(opt.data_path,transform=transforms) dataloader = t.utils.data.DataLoader(dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_workers, drop_last=True) #step2: 定义网络 netg,netd = NetG(opt),NetD(opt) map_location = lambda storage,loc:storage if opt.netd_path: netd.load_state_dict(t.load(opt.netd_path, map_location=map_location)) if opt.netg_path: netg.load_state_dict(t.load(opt.netg_path, map_location=map_location)) #定义优化器和损失函数 optimizer_g = t.optim.Adam(netg.parameters(), opt.lrG, betas=(0.5, 0.999)) optimizer_d = t.optim.Adam(netd.parameters(), opt.lrD, betas=(0.5, 0.999)) criterion = t.nn.BCELoss() #真图片label为1,加图片label为0 #noise为网络的输入 true_labels = t.ones(opt.batch_size) fake_labels = t.zeros(opt.batch_size) fix_noises = t.randn(opt.batch_size,opt.nz,1,1) noises = t.randn(opt.batch_size,opt.nz,1,1) errord_meter = AverageValueMeter() errorg_meter = AverageValueMeter() if opt.gpu: device = t.device("cuda:0" if t.cuda.is_available() else "cpu") netd.to(device) netg.to(device) criterion.to(device) true_labels,fake_labels = true_labels.to(device),fake_labels.to(device) fix_noises,noises = fix_noises.to(device),noises.to(device) epochs = range(140) for epoch in iter(epochs): for ii,(img,_) in tqdm.tqdm(enumerate(dataloader),total=len(dataloader)): if opt.gpu: real_img = img.to(device) if ii%opt.d_every == 0: #每个batch训练一次鉴别器 optimizer_d.zero_grad() output = netd(real_img) #判断真图片(使其尽可能大) error_d_real = criterion(output,true_labels) error_d_real.backward() ##尽可能把假图片判断为错误 noises.data.copy_(t.randn(opt.batch_size,opt.nz,1,1)) fake_img = netg(noises).detach() #根据噪声生成假图 output = netd(fake_img) error_d_fake = criterion(output,fake_labels) error_d_fake.backward() optimizer_d.step() error_d = error_d_fake + error_d_real errord_meter.add(error_d.item()) if ii%opt.g_every == 0: #每5个batch更新一次生成器 #训练生成器 optimizer_g.zero_grad() noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises) output = netd(fake_img) error_g = criterion(output,true_labels) error_g.backward() optimizer_g.step() errord_meter.add(error_g.item()) if opt.vis and ii%opt.plot_time == opt.plot_time - 1: ##可视化 fix_fake_img = netg(fix_noises) #使用噪声生成图片 vis.images(fix_fake_img.data.cpu().numpy()[:64]*0.5+0.5,win = 'fixfake') # vis.images(real_img.data.cpu().numpy()[:64]*0.5+0.5,win = 'real') vis.plot('errord', errord_meter.value()[0]) vis.plot('errorg', errorg_meter.value()[0]) if epoch%opt.decay_every == opt.decay_every-1: #保存模型,图片 tv.utils.save_image(fix_fake_img.data[:64],'%s/new%s.png'%(opt.save_path,epoch), normalize=True,range=(-1,1)) t.save(netd.state_dict(), 'checkpoints/new_netd_%s.pth' % epoch) t.save(netg.state_dict(), 'checkpoints/new_netg_%s.pth' % epoch) errord_meter.reset() errorg_meter.reset() optimizer_g = t.optim.Adam(netg.parameters(), opt.lrG, betas=(0.5, 0.999)) optimizer_d = t.optim.Adam(netd.parameters(), opt.lrD, betas=(0.5, 0.999))
class TACGAN(): def __init__(self, args): self.lr = args.lr self.cuda = args.use_cuda self.batch_size = args.batch_size self.image_size = args.image_size self.epochs = args.epochs self.data_root = args.data_root self.dataset = args.dataset self.save_dir = args.save_dir self.save_prefix = args.save_prefix self.continue_training = args.continue_training self.continue_epoch = args.continue_epoch self.netG_path = args.netg_path self.netD_path = args.netd_path self.save_after = args.save_after self.trainset_loader = None self.evalset_loader = None self.num_workers = args.num_workers self.docvec_size = args.docvec_size self.n_z = args.n_z # length of the noise vector self.nl_d = args.nl_d self.nl_g = args.nl_g self.nf_g = args.nf_g self.nf_d = args.nf_d self.bce_loss = nn.BCELoss() self.nll_loss = nn.NLLLoss() self.mse_loss = nn.MSELoss() self.class_filename = args.class_filename class_path = os.path.join(self.data_root, self.dataset, self.class_filename) with open(class_path) as f: self.num_classes = len([l for l in f]) print(self.num_classes) self.netD = NetD(n_cls=self.num_classes, n_t=self.nl_d, n_f=self.nf_d, docvec_size=self.docvec_size) self.netG = NetG(n_z=self.n_z, n_l=self.nl_g, n_c=self.nf_g, n_t=self.docvec_size) if self.continue_training: self.loadCheckpoints() # convert to cuda tensors if self.cuda and torch.cuda.is_available(): print('CUDA is enabled') self.netD = nn.DataParallel(self.netD).cuda() self.netG = nn.DataParallel(self.netG).cuda() self.bce_loss = self.bce_loss.cuda() self.nll_loss = self.nll_loss.cuda() self.mse_loss = self.mse_loss.cuda() # optimizers for netD and netG self.optimizerD = optim.Adam(params=self.netD.parameters(), lr=self.lr, betas=(0.5, 0.999)) self.optimizerG = optim.Adam(params=self.netG.parameters(), lr=self.lr, betas=(0.5, 0.999)) # create dir for saving checkpoints and other results if do not exist if not os.path.exists(self.save_dir): os.makedirs(self.save_dir) if not os.path.exists(os.path.join(self.save_dir,'netd_checkpoints')): os.makedirs(os.path.join(self.save_dir,'netd_checkpoints')) if not os.path.exists(os.path.join(self.save_dir,'netg_checkpoints')): os.makedirs(os.path.join(self.save_dir,'netg_checkpoints')) if not os.path.exists(os.path.join(self.save_dir,'generated_images')): os.makedirs(os.path.join(self.save_dir,'generated_images')) # start training process def train(self): # write to the log file and print it log_msg = '********************************************\n' log_msg += ' Training Parameters\n' log_msg += 'Dataset:%s\nImage size:%dx%d\n'%(self.dataset, self.image_size, self.image_size) log_msg += 'Batch size:%d\n'%(self.batch_size) log_msg += 'Number of epochs:%d\nlr:%f\n'%(self.epochs,self.lr) log_msg += 'nz:%d\nnl-d:%d\nnl-g:%d\n'%(self.n_z, self.nl_d, self.nl_g) log_msg += 'nf-g:%d\nnf-d:%d\n'%(self.nf_g, self.nf_d) log_msg += '********************************************\n\n' print(log_msg) with open(os.path.join(self.save_dir, 'training_log.txt'),'a') as log_file: log_file.write(log_msg) # load trainset and evalset imtext_ds = ImTextDataset(data_dir=self.data_root, dataset=self.dataset, train=True, image_size=self.image_size) self.trainset_loader = DataLoader(dataset=imtext_ds, batch_size=self.batch_size, shuffle=True, num_workers=self.num_workers) print("Dataset loaded successfuly") # load checkpoints for continuing training # repeat for the number of epochs netd_losses = [] netg_losses = [] for epoch in range(self.epochs): netd_loss, netg_loss = self.trainEpoch(epoch) netd_losses.append(netd_loss) netg_losses.append(netg_loss) self.saveGraph(netd_losses,netg_losses) #self.evalEpoch(epoch) self.saveCheckpoints(epoch) # train epoch def trainEpoch(self, epoch): self.netD.train() # set to train mode self.netG.train() #! set to train mode??? netd_loss_sum = 0 netg_loss_sum = 0 start_time = time() for i, (images, labels, captions) in enumerate(self.trainset_loader): batch_size = images.size(0) # !batch size my be different (from self.batch_size) for the last batch images, labels, captions = Variable(images), Variable(labels), Variable(captions) # !labels should be LongTensor labels = labels.type(torch.FloatTensor) # convert to FloatTensor (from DoubleTensor) lbl_real = Variable(torch.ones(batch_size, 1)) lbl_fake = Variable(torch.zeros(batch_size, 1)) noise = Variable(torch.randn(batch_size, self.n_z)) # create random noise noise.data.normal_(0,1) # normalize the noise rnd_perm1 = torch.randperm(batch_size) # random permutations for different sets of training tuples rnd_perm2 = torch.randperm(batch_size) rnd_perm3 = torch.randperm(batch_size) rnd_perm4 = torch.randperm(batch_size) if self.cuda: images, labels, captions = images.cuda(), labels.cuda(), captions.cuda() lbl_real, lbl_fake = lbl_real.cuda(), lbl_fake.cuda() noise = noise.cuda() rnd_perm1, rnd_perm2, rnd_perm3, rnd_perm4 = rnd_perm1.cuda(), rnd_perm2.cuda(), rnd_perm3.cuda(), rnd_perm4.cuda() ############### Update NetD ############### self.netD.zero_grad() # train with wrong image, wrong label, real caption outD_wrong, outC_wrong = self.netD(images[rnd_perm1], captions[rnd_perm2]) # lossD_wrong = self.bce_loss(outD_wrong, lbl_fake) lossD_wrong = self.bce_loss(outD_wrong, lbl_fake) + self.mse_loss(outD_wrong, lbl_fake) lossC_wrong = self.bce_loss(outC_wrong, labels[rnd_perm1]) # train with real image, real label, real caption outD_real, outC_real = self.netD(images, captions) #lossD_real = self.bce_loss(outD_real, lbl_real) lossD_real = self.bce_loss(outD_real, lbl_real) + self.mse_loss(outD_real, lbl_real) lossC_real = self.bce_loss(outC_real, labels) # train with fake image, real label, real caption fake = self.netG(noise, captions) outD_fake, outC_fake = self.netD(fake.detach(), captions[rnd_perm3]) #lossD_fake = self.bce_loss(outD_fake, lbl_fake) lossD_fake = self.bce_loss(outD_fake, lbl_fake) + self.mse_loss(outD_fake, lbl_fake) lossC_fake = self.bce_loss(outC_fake, labels[rnd_perm3]) # backward and forwad for NetD netD_loss = lossC_wrong+lossC_real+lossC_fake + lossD_wrong+lossD_real+lossD_fake netD_loss.backward() self.optimizerD.step() ########## Update NetG ########## # train with fake data self.netG.zero_grad() noise.data.normal_(0,1) # normalize the noise vector fake = self.netG(noise, captions[rnd_perm4]) d_fake, c_fake = self.netD(fake, captions[rnd_perm4]) #lossD_fake_G = self.bce_loss(d_fake, lbl_real) lossD_fake_G = self.mse_loss(d_fake, lbl_real) lossC_fake_G = self.bce_loss(c_fake, labels[rnd_perm4]) netG_loss = lossD_fake_G + lossC_fake_G netG_loss.backward() self.optimizerG.step() netd_loss_sum += netD_loss.item() netg_loss_sum += netG_loss.item() ### print progress info ### print('Epoch %d/%d, %.2f%% completed. Loss_NetD: %.4f, Loss_NetG: %.4f' %(epoch, self.epochs,(float(i)/len(self.trainset_loader))*100, netD_loss.item(), netG_loss.item())) end_time = time() netd_avg_loss = netd_loss_sum / len(self.trainset_loader) netg_avg_loss = netg_loss_sum / len(self.trainset_loader) epoch_time = (end_time-start_time)/60 log_msg = '-------------------------------------------\n' log_msg += 'Epoch %d took %.2f minutes\n'%(epoch, epoch_time) log_msg += 'NetD average loss: %.4f, NetG average loss: %.4f\n\n' %(netd_avg_loss, netg_avg_loss) print(log_msg) with open(os.path.join(self.save_dir, 'training_log.txt'),'a') as log_file: log_file.write(log_msg) return netd_avg_loss, netg_avg_loss # eval epoch def evalEpoch(self, epoch): #self.netD.eval() #self.netG.eval() return 0 # draws and saves the loss graph upto the current epoch def saveGraph(self, netd_losses, netg_losses): plt.plot(netd_losses, color='red', label='NetD Loss') plt.plot(netg_losses, color='blue', label='NetG Loss') plt.xlabel('epoch') plt.ylabel('error') plt.legend(loc='best') plt.savefig(os.path.join(self.save_dir,'loss_graph.png')) plt.close() # save after each epoch def saveCheckpoints(self, epoch): if epoch%self.save_after==0: name_netD = "netd_checkpoints/netD_" + self.save_prefix + "_epoch_" + str(epoch) + ".pth" name_netG = "netg_checkpoints/netG_" + self.save_prefix + "_epoch_" + str(epoch) + ".pth" torch.save(self.netD.module.state_dict(), os.path.join(self.save_dir, name_netD)) torch.save(self.netG.module.state_dict(), os.path.join(self.save_dir, name_netG)) print("Checkpoints for epoch %d saved successfuly" %(epoch)) # SAVE: data parallel model => add .module # LOAD: create model and load checkpoints(not add .module) and wrap nn.DataParallel # this is for fitting prefix # load checkpoints to continue training def loadCheckpoints(self): name_netD = "netd_checkpoints/netD_" + self.save_prefix + "_epoch_" + str(self.continue_epoch) + ".pth" name_netG = "netg_checkpoints/netG_" + self.save_prefix + "_epoch_" + str(self.continue_epoch) + ".pth" self.netG.load_state_dict(torch.load(os.path.join(self.save_dir, name_netG))) self.netD.load_state_dict(torch.load(os.path.join(self.save_dir, name_netD))) print("Checkpoints loaded successfuly")
netG = NetG(cfg.TRAIN.NF, 100).cuda() netD = DISC_ENCODER_V2(cfg.TRAIN.NF).cuda() text_encoder = RNN_ENCODER(dataset.n_words, nhidden=cfg.TEXT.EMBEDDING_DIM) state_dict = torch.load(cfg.TEXT.DAMSM_NAME, map_location=lambda storage, loc: storage) text_encoder.load_state_dict(state_dict) text_encoder.cuda() for p in text_encoder.parameters(): p.requires_grad = False text_encoder.eval() state_epoch = args.resume_epoch optimizerG = torch.optim.Adam(netG.parameters(), lr=0.0001, betas=(0.0, 0.9)) optimizerD = torch.optim.Adam(netD.parameters(), lr=0.0004, betas=(0.0, 0.9)) if state_epoch != 0: netG.load_state_dict(torch.load('%s/models/netG_%03d.pth' % (output_dir, state_epoch),map_location='cpu')) netD.load_state_dict(torch.load('%s/models/netD_%03d.pth' % (output_dir, state_epoch),map_location='cpu')) netG = netG.cuda() netD = netD.cuda() optimizerG.load_state_dict(torch.load('%s/models/optimizerG.pth' %(output_dir))) optimizerD.load_state_dict(torch.load('%s/models/optimizerD.pth' % (output_dir))) netG.cuda() netD.cuda() if cfg.B_VALIDATION: sampling(text_encoder, netG, dataloader,device) # generate images for the whole valid dataset
torchvision.transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)), ]) dataset = torchvision.datasets.ImageFolder(opt.data_path, transform=transforms) dataloader = torch.utils.data.DataLoader( dataset=dataset, batch_size=opt.batchSize, shuffle=True, drop_last=True, ) netG = NetG(opt.ngf, opt.nz).to(device) netD = NetD(opt.ndf).to(device) criterion = nn.BCELoss() optimizerG = torch.optim.Adam(netG.parameters(), lr=opt.lr, betas=(opt.beta1, 0.999)) optimizerD = torch.optim.Adam(netD.parameters(), lr=opt.lr, betas=(opt.beta1, 0.999)) label = torch.FloatTensor(opt.batchSize) real_label = 1 fake_label = 0 if not os.path.exists(opt.outf): os.makedirs(opt.outf) if not os.path.exists(opt.model_path): os.makedirs(opt.model_path) for epoch in range(1, opt.epoch + 1): for i, (imgs,_) in enumerate(dataloader): optimizerD.zero_grad() imgs=imgs.to(device)
class AnoGAN: """AnoGAN Class """ def __init__(self, opt): # super(AnoGAN, self).__init__(opt, dataloader) # Initalize variables. self.opt = opt self.niter = self.opt.niter self.start_iter = 0 self.netd_niter = 5 self.test_iter = 100 self.lr = self.opt.lr self.batchsize = {'train': self.opt.batchsize, 'test': 1} self.pretrained = False self.phase = 'train' self.outf = self.opt.experiment_group self.algorithm = 'wgan' # LOAD DATA SET self.dataloader = { 'train': provider('train', opt.category, batch_size=self.batchsize['train'], num_workers=4), 'test': provider('test', opt.category, batch_size=self.batchsize['test'], num_workers=4) } self.trn_dir = os.path.join(self.outf, self.opt.experiment_name, 'train') self.tst_dir = os.path.join(self.outf, self.opt.experiment_name, 'test') self.test_img_dir = os.path.join(self.outf, self.opt.experiment_name, 'test', 'images') if not os.path.isdir(self.test_img_dir): os.makedirs(self.test_img_dir) self.best_test_dir = os.path.join(self.outf, self.opt.experiment_name, 'test', 'best_images') if not os.path.isdir(self.best_test_dir): os.makedirs(self.best_test_dir) self.weight_dir = os.path.join(self.trn_dir, 'weights') if not os.path.exists(self.weight_dir): os.makedirs(self.weight_dir) # -- Misc attributes self.epoch = 0 self.l_con = l1_loss self.l_enc = l2_loss ## # Create and initialize networks. self.netg = NetG().cuda() self.netd = NetD().cuda() # Setup optimizer self.optimizer_d = optim.RMSprop(self.netd.parameters(), lr=self.lr) self.optimizer_g = optim.Adam(self.netg.parameters(), lr=self.lr) ## self.weight_path = os.path.join(self.outf, self.opt.experiment_name, 'train', 'weights') if os.path.exists(self.weight_path) and len( os.listdir(self.weight_path)) == 2: print("Loading pre-trained networks...\n") self.netg.load_state_dict( torch.load(os.path.join(self.weight_path, 'netG.pth'))['state_dict']) self.netd.load_state_dict( torch.load(os.path.join(self.weight_path, 'netD.pth'))['state_dict']) self.optimizer_g.load_state_dict( torch.load(os.path.join(self.weight_path, 'netG.pth'))['optimizer']) self.optimizer_d.load_state_dict( torch.load(os.path.join(self.weight_path, 'netD.pth'))['optimizer']) self.start_iter = torch.load( os.path.join(self.weight_path, 'netG.pth'))['epoch'] ## def start(self): """ Train the model """ ## # TRAIN # self.total_steps = 0 best_criterion = -1 #float('inf') best_auc = -1 # Train for niter epochs. # print(">> Training model %s." % self.name) for self.epoch in range(self.start_iter, self.niter): # Train for one epoch mean_wass = self.train() (auc, res, best_rec, best_threshold), res_total = self.test() message = '' # message += 'criterion: (%.3f+%.3f)/2=%.3f ' % (best_rec[0], best_rec[1], res) # message += 'best threshold: %.3f ' % best_threshold message += 'Wasserstein Distance:%.3d ' % mean_wass message += 'AUC: %.3f ' % auc print(message) torch.save( { 'epoch': self.epoch + 1, 'state_dict': self.netg.state_dict(), 'optimizer': self.optimizer_g.state_dict() }, '%s/netG.pth' % (self.weight_dir)) torch.save( { 'epoch': self.epoch + 1, 'state_dict': self.netd.state_dict(), 'optimizer': self.optimizer_d.state_dict() }, '%s/netD.pth' % (self.weight_dir)) if auc > best_auc: best_auc = auc new_message = "******** New optimal found, saving state ********" message = message + '\n' + new_message print(new_message) for img in os.listdir(self.best_test_dir): os.remove(os.path.join(self.best_test_dir, img)) for img in os.listdir(self.test_img_dir): shutil.copyfile(os.path.join(self.test_img_dir, img), os.path.join(self.best_test_dir, img)) shutil.copyfile('%s/netG.pth' % (self.weight_dir), '%s/netg_best.pth' % (self.weight_dir)) log_name = os.path.join(self.outf, self.opt.experiment_name, 'loss_log.txt') message = 'Epoch%3d:' % self.epoch + ' ' + message with open(log_name, "a") as log_file: if self.epoch == 0: log_file.write('\n\n') log_file.write('%s\n' % message) print(">> Training %s Done..." % self.opt.experiment_name) ## def train(self): """ Train the model for one epoch. """ print("\n>>> Epoch %d/%d, Running " % (self.epoch + 1, self.niter) + self.opt.experiment_name) self.netg.train() self.netd.train() # for p in self.netg.parameters(): p.requires_grad = True mean_wass = 0 tk0 = tqdm(self.dataloader['train'], total=len(self.dataloader['train'])) for i, itr in enumerate(tk0): input, _ = itr input = input.cuda() wasserstein_d = None # if self.algorithm == 'wgan': # train NetD for _ in range(self.netd_niter): # for p in self.netd.parameters(): p.requires_grad = True self.optimizer_d.zero_grad() # forward_g latent_i = torch.rand(self.batchsize['train'], 64, 1, 1).cuda() fake = self.netg(latent_i) # forward_d _, pred_real = self.netd(input) _, pred_fake = self.netd(fake) # .detach() TODO # Backward-pass wasserstein_d = (pred_fake.mean() - pred_real.mean()) * 1 wasserstein_d.backward() self.optimizer_d.step() for p in self.netd.parameters(): p.data.clamp_(-0.01, 0.01) #<<<<<<< # train netg # for p in self.netd.parameters(): p.requires_grad = False self.optimizer_g.zero_grad() noise = torch.rand(self.batchsize['train'], 64, 1, 1).cuda() fake = self.netg(noise) _, pred_fake = self.netd(fake) err_g_d = -pred_fake.mean() # negative err_g_d.backward() self.optimizer_g.step() errors = { 'loss_netD': wasserstein_d.item(), 'loss_netG': round(err_g_d.item(), 3), } mean_wass += wasserstein_d.item() tk0.set_postfix(errors) if i % 50 == 0: img_dir = os.path.join(self.outf, self.opt.experiment_name, 'train', 'images') if not os.path.isdir(img_dir): os.makedirs(img_dir) self.save_image_cv2(input.data, '%s/reals.png' % img_dir) self.save_image_cv2(fake.data, '%s/fakes%03d.png' % (img_dir, i)) mean_wass /= len(self.dataloader['train']) return mean_wass ## def test(self): """ Test AnoGAN model. Args: dataloader ([type]): Dataloader for the test set Raises: IOError: Model weights not found. """ self.netg.eval() self.netd.eval() # for p in self.netg.parameters(): p.requires_grad = False # for p in self.netd.parameters(): p.requires_grad = False for img in os.listdir(self.test_img_dir): os.remove(os.path.join(self.test_img_dir, img)) self.phase = 'test' meter = Meter_AnoGAN() tk1 = tqdm(self.dataloader['test'], total=len(self.dataloader['test'])) for i, itr in enumerate(tk1): input, target = itr input = input.cuda() latent_i = torch.rand(self.batchsize['test'], 64, 1, 1).cuda() latent_i.requires_grad = True optimizer_latent = optim.Adam([latent_i], lr=self.lr) test_loss = None for _ in range(self.test_iter): optimizer_latent.zero_grad() fake = self.netg(latent_i) residual_loss = self.l_con(input, fake) latent_o, _ = self.netd(fake) discrimination_loss = self.l_enc(latent_i, latent_o) alpha = 0.1 test_loss = ( 1 - alpha) * residual_loss + alpha * discrimination_loss test_loss.backward() optimizer_latent.step() abnormal_score = test_loss meter.update(abnormal_score, target) #<<<TODO # Save test images. combine = torch.cat([input.cpu(), fake.cpu()], dim=0) self.save_image_cv2(combine, '%s/%05d.jpg' % (self.test_img_dir, i + 1)) criterion, res_total = meter.get_metrics() # rename images for i, res in enumerate(res_total): os.rename('%s/%05d.jpg' % (self.test_img_dir, i + 1), '%s/%05d_%s.jpg' % (self.test_img_dir, i + 1, res)) return criterion, res_total @staticmethod def save_image_cv2(tensor, filename): # return from torchvision.utils import make_grid # tensor = (tensor + 1) / 2 grid = make_grid(tensor, 8, 2, 0, False, None, False) ndarray = grid.mul_(255).clamp_(0, 255).permute(1, 2, 0).to( 'cpu', torch.uint8).numpy() cv2.imwrite(filename, ndarray)
netg = netg.to(device) netd = netd.to(device) neta = neta.to(device) netg.train() netd.train() neta.train() dataset = CASIABDataset(data_dir='../data/GEI_CASIA_B/gei/') iteration = 0 lr = 0.0002 real_label = 1 fake_label = 0 fineSize = 64 label = th.zeros((128, 1), requires_grad=False).to(device) optimG = optim.Adam(netg.parameters(), lr=lr/2) optimD = optim.Adam(netd.parameters(), lr=lr/3) optimA = optim.Adam(neta.parameters(), lr=lr/3) print('Training starts') while iteration < 1000000: ass_label, noass_label, img = dataset.getbatch(128) ass_label = ass_label.to(device).to(th.float32) noass_label = noass_label.to(device).to(th.float32) img = img.to(device).to(th.float32) # update D lossD = 0 optimD.zero_grad() output = netd(ass_label) label.fill_(real_label) lossD_real1 = F.binary_cross_entropy(output, label) lossD += lossD_real1.item()
netG = NetG(cfg.TRAIN.NF, 100).cuda() netD = NetProjD_V2().cuda() text_encoder = RNN_ENCODER(dataset.n_words, nhidden=cfg.TEXT.EMBEDDING_DIM) state_dict = torch.load(cfg.TEXT.DAMSM_NAME, map_location=lambda storage, loc: storage) text_encoder.load_state_dict(state_dict) text_encoder.cuda() for p in text_encoder.parameters(): p.requires_grad = False text_encoder.eval() state_epoch = args.resume_epoch optimizerG = torch.optim.Adam(netG.parameters(), lr=0.0001, betas=(0.0, 0.9)) optimizerD_enc = torch.optim.Adam(netD.feature_encoder.parameters(), lr=0.0004, betas=(0.0, 0.9)) optimizerD_proj = torch.optim.Adam(netD.COND_DNET.parameters(), lr=0.004, betas=(0.0, 0.9)) if state_epoch != 0: netG.load_state_dict( torch.load('%s/models/netG_%03d.pth' % (output_dir, state_epoch), map_location='cpu')) netD.load_state_dict( torch.load('%s/models/netD_%03d.pth' % (output_dir, state_epoch),
def train(**kwargs): # 读取参数赋值 for k_, v_ in kwargs.items(): setattr(opt, k_, v_) # 可视化 if opt.vis: from visualize import Visualizer vis = Visualizer(opt.env) # 对图片进行操作 transforms = tv.transforms.Compose([ tv.transforms.Scale(opt.image_size), tv.transforms.CenterCrop(opt.image_size), tv.transforms.ToTensor(), # 均值方差 tv.transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)) ]) # ImageFolder 使用pytorch原生的方法读取图片,并进行操作 封装数据集 dataset = tv.datasets.ImageFolder(opt.data_path, transform=transforms) #数据加载器 dataloader = t.utils.data.DataLoader(dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_workers, drop_last=True) # 定义网络 netg, netd = NetG(opt), NetD(opt) # 把map内容加载到CPU中 map_location = lambda storage, loc: storage # 将预训练的模型都先加载到cpu上 if opt.netd_path: netd.load_state_dict(t.load(opt.netd_path, map_location=map_location)) if opt.netg_path: netg.load_state_dict(t.load(opt.netg_path, map_location=map_location)) # 定义优化器和损失 optimizer_g = t.optim.Adam(netg.parameters(), opt.lr1, betas=(opt.beta1, 0.999)) optimizer_d = t.optim.Adam(netd.parameters(), opt.lr2, betas=(opt.beta1, 0.999)) # BinaryCrossEntropy二分类交叉熵,常用于二分类问题,当然也可以用于多分类问题 criterion = t.nn.BCELoss() # 真图片label为1,假图片label为0 # noises为生成网络的输入 true_labels = Variable(t.ones(opt.batch_size)) fake_labels = Variable(t.zeros(opt.batch_size)) # fix_noises是固定值,用来查看每个epoch的变化效果 fix_noises = Variable(t.randn(opt.batch_size, opt.nz, 1, 1)) noises = Variable(t.randn(opt.batch_size, opt.nz, 1, 1)) # AverageValueMeter统计任意添加的变量的方差和均值 可视化的仪表盘 errord_meter = AverageValueMeter() errorg_meter = AverageValueMeter() if opt.gpu: # 网络转移到GPU netd.cuda() netg.cuda() # 损失函数转移到GPU criterion.cuda() # 标签转移到GPU true_labels, fake_labels = true_labels.cuda(), fake_labels.cuda() # 输入噪声转移到GPU fix_noises, noises = fix_noises.cuda(), noises.cuda() epochs = range(opt.max_epoch) for epoch in iter(epochs): for ii, (img, _) in tqdm.tqdm(enumerate(dataloader)): real_img = Variable(img) if opt.gpu: real_img = real_img.cuda() # 每d_every个batch训练判别器 if ii % opt.d_every == 0: # 训练判别器 optimizer_d.zero_grad() ## 尽可能的把真图片判别为正确 #一个batchd的真照片判定为1 并反向传播 output = netd(real_img) error_d_real = criterion(output, true_labels) #反向传播 error_d_real.backward() ## 尽可能把假图片判别为错误 # 一个batchd的假照片判定为0 并反向传播 noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) fake_img = netg(noises).detach() # 根据噪声生成假图 output = netd(fake_img) error_d_fake = criterion(output, fake_labels) error_d_fake.backward() #更新可学习参数 optimizer_d.step() # 总误差=识别真实图片误差+假图片误差 error_d = error_d_fake + error_d_real # 将总误差加入仪表板用于可视化显示 errord_meter.add(error_d.data[0]) # 每g_every个batch训练生成器 if ii % opt.g_every == 0: # 训练生成器 optimizer_g.zero_grad() noises.data.copy_(t.randn(opt.batch_size, opt.nz, 1, 1)) # 生成器:噪声生成假图片 fake_img = netg(noises) # 判别器:假图片判别份数 output = netd(fake_img) # 尽量让假图片的份数与真标签接近,让判别器分不出来 error_g = criterion(output, true_labels) error_g.backward() # 更新参数 optimizer_g.step() # 将误差加入仪表板用于可视化显示 errorg_meter.add(error_g.data[0]) if opt.vis and ii % opt.plot_every == opt.plot_every - 1: ## 可视化 # 进入debug模式 if os.path.exists(opt.debug_file): ipdb.set_trace() # 固定噪声生成假图片 fix_fake_imgs = netg(fix_noises) # 可视化 固定噪声产生的假图片 vis.images(fix_fake_imgs.data.cpu().numpy()[:64] * 0.5 + 0.5, win='fixfake') # 可视化一张真图片。作为对比 vis.images(real_img.data.cpu().numpy()[:64] * 0.5 + 0.5, win='real') # 可视化仪表盘 判别器误差 生成器误差 vis.plot('errord', errord_meter.value()[0]) vis.plot('errorg', errorg_meter.value()[0]) # 每decay_every个epoch之后保存一次模型 if epoch % opt.decay_every == 0: # 保存模型、图片 tv.utils.save_image(fix_fake_imgs.data[:64], '%s/%s.png' % (opt.save_path, epoch), normalize=True, range=(-1, 1)) # 保存判别器 生成器 t.save(netd.state_dict(), 'checkpoints/netd_%s.pth' % epoch) t.save(netg.state_dict(), 'checkpoints/netg_%s.pth' % epoch) # 清空误差仪表盘 errord_meter.reset() errorg_meter.reset() # 重置优化器参数为刚开始的参数 optimizer_g = t.optim.Adam(netg.parameters(), opt.lr1, betas=(opt.beta1, 0.999)) optimizer_d = t.optim.Adam(netd.parameters(), opt.lr2, betas=(opt.beta1, 0.999))