About the train.py

[Qijian-Z]

Hello, let me ask the question again. When I run the train.py, I got an error message called"RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor". And I tried to fix it, but couldn't.

[chunbaobao]

Alright, I just realized that I forgot to transfer the input tensor of test dataloader to GPU when the parallel is enable which can lead to the different devices error. The new implementation has been committed. By the way, if u only have one or no GPU, u can disable the parallel parameter.

[Qijian-Z]

Thanks a lot!

[Qijian-Z]

Sorry, I have one more question. I tried to run the train.py using ImageNet Dataset, it stopped here. Training Start loading data of imagenet Namespace(seed=2048, lr=0.001, epochs=300, batch_size=32, weight_decay=0.0005, channel='AWGN', saved='./saved', snr_list=[19.0, 13.0, 7.0, 4.0, 1.0], ratio_list=[0.3333333333333333, 0.16666666666666666, 0.08333333333333333], num_workers=4, dataset='imagenet', parallel=True, if_scheduler=False, step_size=640, device='cuda:0', gamma=0.5, disable_tqdm=True) the inner channel is 19 What should I do?

[chunbaobao]

Actually, the ImageNet dataset is quite extensive, so it appears that the model has been initiated, but the first epoch has not been completed yet. u can modify the disable_tqdm option in the arguments to enable the visualization of the training progress.

[Qijian-Z]

I would like to know about the results, but after evaluating the model, I feel it is not performing well. I would like to ask, did you actually run the code and get results similar to the paper?

[chunbaobao]

could you show me your details of parameter and the results? @Qijian-Z

[Qijian-Z]

Alright, I used the model named " imagenet_100_0.33_100.00_32_19.pth ", snr = 20, ratio = 1/3, got the psnr = 13.01. I may have misunderstood, doesn't 100.00 mean you trained under snr = 100?

[Luckwjf]

The project you shared is very good and useful. Thank you very much for your efforts. We look forward to more project sharing in semantic communication.

[chunbaobao]

Alright, I used the model named " imagenet_100_0.33_100.00_32_19.pth ", snr = 20, ratio = 1/3, got the psnr = 13.01. I may have misunderstood, doesn't 100.00 mean you trained under snr = 100?

Yes, it is. The repository is simply for a sanity check, so the visualization.py file has been excluded.

[chunbaobao]

The project you shared is very good and useful. Thank you very much for your efforts. We look forward to more project sharing in semantic communication.

Thanks u so much, I will keep working to share.

[Qijian-Z]

Thank u for your reply. I'm so sorry, could u provide us the visualization.py file?

[chunbaobao]

Thank u for your reply. I'm so sorry, could u provide us the visualization.py file?

sorry, im busy resently. The visualization of the results hasn't been written yet, but it is on my to-do list.

[Qijian-Z]

Thank u. please tell me if you wrote it. Respect.

[chunbaobao]

Updated. The results of cifar10 dataset in AWGN channel are provided in README.md/results.