hulianyuyy / sen_cslr Goto Github PK

Hello，I want to ask, I want to use your trained weights to test my own video data, how to achieve. Is there a demo, please?

Was the Spatial Attention Network, such as SENet or ECA-Net as mentioned in the paper, tried as a backbone before implementing SEN_CSLR? If so, did the results improve?

An error is reported after executing this naming.

The code implementation of your TSEM module is not the same as in the paper, can you disclose the same code part as the paper, thank you~

Hi, when I run test_one_video.py using CSL-Daily pretrained model (dev_31.10_CSL-Daily.pt), I meet the following error

RuntimeError: Error(s) in loading state_dict for SLRModel:
size mismatch for conv1d.fc.weight: copying a param with shape torch.Size([2001, 1024]) from checkpoint, the shape in current model is torch.Size([1296, 1024]).
size mismatch for conv1d.fc.bias: copying a param with shape torch.Size([2001]) from checkpoint, the shape in current model is torch.Size([1296]).
size mismatch for classifier.weight: copying a param with shape torch.Size([2001, 1024]) from checkpoint, the shape in current model is torch.Size([1296, 1024]).
size mismatch for classifier.bias: copying a param with shape torch.Size([2001]) from checkpoint, the shape in current model is torch.Size([1296]).

Should I modify somewhere of the network structure?
Thank you!

Hello, I have trained SEN-CSLR on the data set of CSL-Daily, but WER has not reached 31.1% and 30.7% as mentioned in the article, which is about 1% short. May I ask which part needs to be modified?

hello,your paper's innovation point is heatmap，why not show this part in the release code, thanks.

hi, me again, may you upload CSL dataset Trained checkpoint, thanks

as title, maybe Baidu Netdisk link， thanks a lot.

Hello How Are You guys

Can This Code Run on Cpu ???

Hello! The following issues were encountered during the training process. Could you give me some guidance？