PaddleGAN中MPRNet训练问题

[ONEISALL-h]

问题描述：
使用PaddleGAN中mprnet_deblurring.yaml进行训练，选用了GoPro数据集，未修改其余参数。
使用环境：
AI Studio在线平台、PaddlePaddle2.2.0、Python3.7
代码如下：

从github上克隆PaddleGAN代码
!git clone https://gitee.com/paddlepaddle/PaddleGAN
安装所需安装包
%cd PaddleGAN/
!pip install -r requirements.txt
!pip install imageio-ffmpeg
%cd applications/
单卡训练
!python -u PaddleGAN/tools/main.py --config-file work/mprnet_deblurring.yaml

报错如下：

[11/18 15:19:07] ppgan INFO: Configs: {'total_iters': 100000, 'output_dir': 'output_dir/mprnet_deblurring-2021-11-18-15-19', 'model': {'name': 'MPRModel', 'generator': {'name': 'MPRNet'}, 'char_criterion': {'name': 'CharbonnierLoss'}, 'edge_criterion': {'name': 'EdgeLoss'}}, 'dataset': {'train': {'name': 'MPRTrain', 'rgb_dir': 'GoPro', 'num_workers': 16, 'batch_size': 4, 'img_options': {'patch_size': 256}}, 'test': {'name': 'MPRTrain', 'rgb_dir': 'GoPro', 'num_workers': 16, 'batch_size': 4, 'img_options': {'patch_size': 256}}}, 'lr_scheduler': {'name': 'CosineAnnealingRestartLR', 'learning_rate': 0.0002, 'periods': [25000, 25000, 25000, 25000], 'restart_weights': [1, 1, 1, 1], 'eta_min': 1e-06}, 'validate': {'interval': 10, 'save_img': False, 'metrics': {'psnr': {'name': 'PSNR', 'crop_border': 4, 'test_y_channel': True}, 'ssim': {'name': 'SSIM', 'crop_border': 4, 'test_y_channel': True}}}, 'optimizer': {'name': 'Adam', 'net_names': ['generator'], 'beta1': 0.9, 'beta2': 0.999, 'epsilon': 1e-08}, 'log_config': {'interval': 10, 'visiual_interval': 5000}, 'snapshot_config': {'interval': 5000}, 'is_train': True, 'profiler_options': None, 'timestamp': '-2021-11-18-15-19'}
W1118 15:19:07.026855 714 device_context.cc:404] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 10.1, Runtime API Version: 10.1
W1118 15:19:07.031368 714 device_context.cc:422] device: 0, cuDNN Version: 7.6.
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/tensor/creation.py:125: DeprecationWarning: np.object is a deprecated alias for the builtin object. To silence this warning, use object by itself. Doing this will not modify any behavior and is safe.
Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
if data.dtype == np.object:
Traceback (most recent call last):
File "PaddleGAN/tools/main.py", line 56, in
main(args, cfg)
File "PaddleGAN/tools/main.py", line 46, in main
trainer.train()
File "/home/aistudio/PaddleGAN/ppgan/engine/trainer.py", line 190, in train
self.model.train_iter(self.optimizers)
File "/home/aistudio/PaddleGAN/ppgan/models/mpr_model.py", line 65, in train_iter
loss_edge.append(self.edge_criterion(restored[i], self.target))
File "/home/aistudio/PaddleGAN/ppgan/models/criterions/pixel_loss.py", line 270, in call
loss = self.loss(self.laplacian_kernel(x), self.laplacian_kernel(y))
File "/home/aistudio/PaddleGAN/ppgan/models/criterions/pixel_loss.py", line 264, in laplacian_kernel
new_filter[:,:,::2,::2] = down*4 # upsample
ValueError: (InvalidArgument) Leaf Tensor (dygraph_tmp_1341) that doesn't stop gradient can't use inplace strategy.
[Hint: Expected self->IsLeaf() && !self->OverridedStopGradient() == false, but received self->IsLeaf() && !self->OverridedStopGradient():1 != false:0.] (at /paddle/paddle/fluid/pybind/imperative.cc:718)

[lyl120117]

非常感谢您使用PaddleGAN，MPRNet训练还没有在最新Paddle版本上验证过，针对这个报错问题，我后面会重点看一下，尽快能够支持训练，谢谢！

[ONEISALL-h]

非常感谢您使用PaddleGAN，MPRNet训练还没有在最新Paddle版本上验证过，针对这个报错问题，我后面会重点看一下，尽快能够支持训练，谢谢！

现在有可支持训练的版本吗

[LielinJiang]

我们尽快修复一下