def train(args):
# parse config
place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
with fluid.dygraph.guard(place):
config = parse_config(args.config)
train_config = merge_configs(config, 'train', vars(args))
print_configs(train_config, 'Train')
# train_model = TSN1.TSNResNet('TSN', train_config['MODEL']['num_layers'],
# train_config['MODEL']['num_classes'],
# train_config['MODEL']['seg_num'], 0.00002)
train_model = ResNet3D.ResNet3D('res3d', train_config['MODEL']['num_layers'],
train_config['MODEL']['num_classes'],
train_config['MODEL']['seg_num'], 0.00002)
opt = fluid.optimizer.Momentum(0.001, 0.9, parameter_list=train_model.parameters())
if args.pretrain:
# 加载上一次训练的模型，继续训练
model, _ = fluid.dygraph.load_dygraph(args.save_dir + '/res3d_model')
train_model.load_dict(model)
# build model
if not os.path.exists(args.save_dir):
os.makedirs(args.save_dir)
# get reader
train_config.TRAIN.batch_size = train_config.TRAIN.batch_size
train_reader = KineticsReader(args.model_name.upper(), 'train', train_config).create_reader()
epochs = args.epoch or train_model.epoch_num()
for i in range(epochs):
for batch_id, data in enumerate(train_reader()):
dy_x_data = np.array([x[0] for x in data]).astype('float32')
y_data = np.array([[x[1]] for x in data]).astype('int64')
img = fluid.dygraph.to_variable(dy_x_data)
# print(img.shape) ##################测试
label = fluid.dygraph.to_variable(y_data)
label.stop_gradient = True
out, acc = train_model(img, label)
loss = fluid.layers.cross_entropy(out, label)
avg_loss = fluid.layers.mean(loss)
avg_loss.backward()
opt.minimize(avg_loss)
train_model.clear_gradients()
if batch_id % 12 == 0:
logger.info("Loss at epoch {} step {}: {}, acc: {}".format(i, batch_id, avg_loss.numpy(), acc.numpy()))
print("Loss at epoch {} step {}: {}, acc: {}".format(i, batch_id, avg_loss.numpy(), acc.numpy()))
fluid.dygraph.save_dygraph(train_model.state_dict(), args.save_dir + '/res3d_model')
logger.info("Final loss: {}".format(avg_loss.numpy()))
print("Final loss: {}".format(avg_loss.numpy()))
def eval(args):
# parse config
config = parse_config(args.config)
val_config = merge_configs(config, 'valid', vars(args))
print_configs(val_config, "Valid")
with fluid.dygraph.guard():
val_model = ResNet3D.ResNet3D('resnet',
val_config['MODEL']['num_layers'],
val_config['MODEL']['num_classes'],
val_config['MODEL']['seg_num'], 0.00002)
label_dic = np.load('label_dir.npy', allow_pickle=True).item()
# print('eval.py 中的 label_dic : ', label_dic)
label_dic = {v: k for k, v in label_dic.items()}
# get infer reader
val_reader = KineticsReader(args.model_name.upper(), 'valid',
val_config).create_reader()
# if no weight files specified, exit()
if args.weights:
weights = args.weights
else:
print("model path must be specified")
exit()
para_state_dict, _ = fluid.load_dygraph(weights)
val_model.load_dict(para_state_dict)
val_model.eval()
acc_list = []
for batch_id, data in enumerate(val_reader()):
dy_x_data = np.array([x[0] for x in data]).astype('float32')
y_data = np.array([[x[1]] for x in data]).astype('int64')
img = fluid.dygraph.to_variable(dy_x_data)
label = fluid.dygraph.to_variable(y_data)
label.stop_gradient = True
out, acc = val_model(img, label)
acc_list.append(acc.numpy()[0])
print("验证集准确率为:{}".format(np.mean(acc_list)))
def infer(args):
# parse config
config = parse_config(args.config)
infer_config = merge_configs(config, 'infer', vars(args))
print_configs(infer_config, "Infer")
with fluid.dygraph.guard():
infer_model = ResNet3D.ResNet3D('res3d',
infer_config['MODEL']['num_layers'],
infer_config['MODEL']['num_classes'],
infer_config['MODEL']['seg_num'],
0.00002)
label_dic = np.load('label_dir.npy', allow_pickle=True).item()
label_dic = {v: k for k, v in label_dic.items()}
# get infer reader
infer_reader = KineticsReader(args.model_name.upper(), 'infer',
infer_config).create_reader()
# if no weight files specified, exit()
if args.weights:
weights = args.weights
else:
print("model path must be specified")
exit()
para_state_dict, _ = fluid.load_dygraph(weights)
infer_model.load_dict(para_state_dict)
infer_model.eval()
for batch_id, data in enumerate(infer_reader()):
dy_x_data = np.array([x[0] for x in data]).astype('float32')
y_data = [x[1] for x in data]
img = fluid.dygraph.to_variable(dy_x_data)
out = infer_model(img).numpy()[0]
label_id = np.where(out == np.max(out))
print("实际标签{}, 预测结果{}".format(y_data, label_dic[label_id[0][0]]))
pin_memory=par.pin_mem)
print("Number of samples in test dataset: %d\n" % len(test_df.index))
#create model
if par.model == "deepvo":
model = DeepVO(par.img_h, par.img_w, par.batch_norm)
elif par.model == "resnet3d":
if par.resnet_depth == 18:
num_blocks = [2, 2, 2, 2]
elif par.resnet_depth == 34:
num_blocks = [3, 4, 6, 3]
else:
raise NotImplementedError("Invalid choice of Resnet depth!")
model = ResNet3D(num_blocks)
else:
raise NotImplementedError("Invalid model selected!")
use_cuda = torch.cuda.is_available()
if use_cuda:
print("Moving model to GPU...\n")
model = model.cuda()
if not par.load_weights or not os.path.isfile(par.load_model_path):
raise ValueError("Pretrained weights not provided!")
else:
model.load_state_dict(torch.load(par.load_model_path))
print("Begin evaluating model...\n")
def train(args):
# parse config 参数配置
place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace() # 是否使用 GPU
with fluid.dygraph.guard(place):
config = parse_config(args.config)
train_config = merge_configs(config, 'train', vars(args)) # vars 函数，返回参数表达式的值
print_configs( train_config, 'Train configs : ' )
train_model = ResNet3D.ResNet3D('resnet',train_config['MODEL']['num_layers'],
train_config['MODEL']['num_classes'],
train_config['MODEL']['seg_num'],
0.00002)
#根据自己定义的网络，声明train_model
# parameter_list 指明在训练的时候，哪些参数( 在此是 train_model.parameters() )会被优化
opt = fluid.optimizer.Momentum(0.001, 0.9, parameter_list=train_model.parameters())
if args.pretrain:
# 加载上一次训练的模型，继续训练
model, _ = fluid.dygraph.load_dygraph(args.save_dir + '/resnet_model')
train_model.load_dict(model)
# 创建一个保存模型的路径
if not os.path.exists(args.save_dir):
os.makedirs(args.save_dir)
# get reader
train_config.TRAIN.batch_size = train_config.TRAIN.batch_size # 两边完全一样啊？？？
# KineticsReader().create_reader() 函数返回值是 batch_size 组 <img, label> 数据
train_reader = KineticsReader(args.model_name.upper(), 'train', train_config).create_reader()
epochs = args.epoch or train_model.epoch_num()
for i in range(epochs):
for batch_id, data in enumerate(train_reader()):
dy_x_data = np.array([x[0] for x in data]).astype('float32')
y_data = np.array([[x[1]] for x in data]).astype('int64')
## 获取的img 是一个5维数据：batchbatch_size,提取多少片段(seg_num*seg_len)，通道数，长，宽
img = fluid.dygraph.to_variable(dy_x_data)
label = fluid.dygraph.to_variable(y_data)
label.stop_gradient = True
out, acc = train_model(img, label)
loss = fluid.layers.cross_entropy(out, label)
avg_loss = fluid.layers.mean(loss)
avg_loss.backward()
opt.minimize(avg_loss)
train_model.clear_gradients()
# 隔多少次训练，进行一次输出提示
if batch_id % 1 == 0:
logger.info("Loss at epoch {} step {}: {}, acc: {}".format(i, batch_id, avg_loss.numpy(), acc.numpy()))
print("Loss at epoch {} step {}: {}, acc: {}".format(i, batch_id, avg_loss.numpy(), acc.numpy()))
fluid.dygraph.save_dygraph(train_model.state_dict(), args.save_dir + '/resnet_model')
logger.info("Final loss: {}".format(avg_loss.numpy()))
print("Final loss: {}".format(avg_loss.numpy()))