How to slove the datalaoder problem when training the model?

[Sranc3]

When I try to use the officile code to train the mdoel, I got the problem when laoding the training data:

Original Traceback (most recent call last):
File "/home/mayue/miniconda3/envs/ADAPT/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py", line 302, in worker_loop
data = fetcher.fetch(index)
File "/home/mayue/miniconda3/envs/ADAPT/lib/python3.8/site-packages/torch/utils/data/utils/fetch.py", line 58, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/mayue/miniconda3/envs/ADAPT/lib/python3.8/site-packages/torch/utils/data/utils/fetch.py", line 58, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/csr/ADAPT/src/datasets/vision_language_tsv.py", line 395, in getitem
raw_frames, is_video = self.get_visual_data(img_idx, start, end)
File "/home/csr/ADAPT/src/datasets/vision_language_tsv.py", line 322, in get_visual_data
row = self.get_row_from_tsv(self.visual_tsv, idx)
File "/home/csr/ADAPT/src/datasets/vision_language_tsv.py", line 171, in get_row_from_tsv
assert row[0].split('/')[0] == self.image_keys[img_idx].split('')[-1] or row[0].split('')[-1] == self.image_keys[img_idx].split('')[-1]
AssertionError

I have seen the original code and I guess you have faced with the same problem, could you tell me how to fix it, thank you very much!

[jxbbb]

Thanks for your interests. I have trained the model again but do not find the bugs. I print some examples and hope this can help you.

[Sranc3]

Thanks for your reply, It may be something wrong with my local train_info files, we will check it carefuly. Thanks again!

[Sranc3]

Thanks for your reply, It may be something wrong with my local train_info files, we will check it carefuly. Thanks again!

I have finally figured out that it's the problem with the download file 'training_32frames_img_size256.img.tsv', it works well in test dataloader and val datalaoder, but the some of the contents are missing in training tsv, and I get a blank row.

[jxbbb]

It seems that some files in dataset or dataset_part are missed. You can check the data structure in https://github.com/jxbbb/ADAPT#dataset-preparation. Note that dataset and dataset_part are both needed when training.

[Sranc3]

Sorry to bother you again, we have redownloaded the code and data, but sitll we get the following problem:

It seems that the gt_signals(torch.size(715)) and pred_signals(torch.size(715,1)) have the wrong shape, so what are they supposed to be?

[jxbbb]

Maybe you can pull the latest code in the hub to your local directory. If this doesn't work, please let me know and I will clone the codes myself and go through the whole process again.

[Sranc3]

Yes, we are using the latest code, and we don't make any changes.