yizhou-li-cv / ecnet Goto Github PK

Thank you very much for sharing!
I'm sorry to bother you, because my foundation is really not very good, and I encountered several problems when running this code. I would really appreciate it if you could give me some guidance!

Fistly, in the code of "python train.py --dataroot./data/SPA-Data_6385_patch --name < autoencoder_path> --model autoencoder_train --dataset_mode rain100h --preprocess none --n_epochs 100 --lr_policy multistep Where gradient_clipping 5", does "autoencoder_path" refer to the path of the "autoencoder_train_model.py" file?

Secondly, if I'm right about the first question, I revised "autoencoder_path" to the path of the "autoencoder_train_model.py" file. Then I encountered this error: ModuleNotFoundError: No module named 'models. deraining. derainlcn_model'.
In the deraining folder, I do not find the file "derainlcn_model". Is it "RLCN.py"?

Wish you good health and smooth work during the epidemic! Looking forward to your reply!

Hello, when I trained the rain100L dataset, the psnr value and ssim value have always been small, what is the reason why it does not match the data in the paper?

Hello,I see the training frameworks of visualizer_training and train are different. Can these two codes be used at the same time? If I use pycharm training it, the command should be same as the readme showing?

Training ECNet would raise an error when tries loading the pre-trained autoencoder:

Unexpected key(s) in state_dict: "Up4.up.1.weight", "Up4.up.1.bias", "Up4.up.2.weight", "Up4.up.2.bias"..... 

It seems that the module is saved in DataParallel when trains with GPU, which stores the model in module, and is trying to load it without DataParallel.

# code snippet from build_networks.py
def init_net(net, init_type='normal', init_gain=0.02, gpu_ids=[], init=True):
    if len(gpu_ids) > 0:
        ...
        net = torch.nn.DataParallel(net, gpu_ids)  # wrapped by DataParallel

# code snippet from base_model.py
def load_networks(self):
   ...
   if isinstance(net, torch.nn.DataParallel):
      net = net.module  # unwrapped as nn.Module

   state_dict = torch.load(load_path, map_location=str(self.device))  # the pre-trained network, save with DataParallel
   net.load_state_dict(model_dict)  # error happens here

I want to know where have used RLCN in the code？I want to receive you reply! Thank you very much!