The train file

[HaFred]

Hi, is it possible to provide a sample train_file? https://github.com/EmbodiedGPT/EmbodiedGPT_Pytorch/blob/93bc120974e4c665894852c304381e42a2a63955/align.sh#L28

Thank you very much for open-sourcing this work. Good job!

[EmbodiedGPT]

we will provide a example train_file and update the readme document as soon as possible

[HaFred]

we will provide a example train_file and update the readme document as soon as possible

Thank you very much. Before the release, is it possible for you to leave a pointer here for the format of the train_file? Like what are the keys text_file and data_name, etc.? Thanks again.

[HaFred]

Thanks for the update @EmbodiedGPT . Could you also release the Ego4D video entries list for https://github.com/EmbodiedGPT/EmbodiedGPT_Pytorch/blob/main/datasets/datasets_share.zip? Maybe something like this?

[EmbodiedGPT]

Due to the fact that the original EGO4D videos are not of very high quality and are difficult to store in large volumes, we have gone through several rounds of cleaning, and therefore do not provide an entries list for the original dataset. Data processing from raw data to slicing of video clips to matching of captions can take months, so we offer a simple and easy-to-use version. We provide a video segmentation after 8 keyframes saved as numpy data and text pairs, base version at https://github.com/EmbodiedGPT/EgoCOT_Dataset, full version at https://huggingface.co/datasets/wofmanaf/ego4d-video.

[HaFred]

Due to the fact that the original EGO4D videos are not of very high quality and are difficult to store in large volumes, we have gone through several rounds of cleaning, and therefore do not provide an entries list for the original dataset. Data processing from raw data to slicing of video clips to matching of captions can take months, so we offer a simple and easy-to-use version. We provide a video segmentation after 8 keyframes saved as numpy data and text pairs, base version at https://github.com/EmbodiedGPT/EgoCOT_Dataset, full version at https://huggingface.co/datasets/wofmanaf/ego4d-video.

I see. When I try to load the model at https://drive.google.com/file/d/1sBTy8oXeweJg3STbhzBR_5pLcVs1F20q/view?usp=sharing by running demo/test.py, the model cannot be corrected loaded in at the following line https://github.com/huggingface/transformers/blob/75b76a5ea461ace0d141d3415879439ae9bbfc22/src/transformers/modeling_utils.py#L533.

Errors: RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory. I tried to load the bin file directly torch.load(), the same error is shown. Would you please check whether the bin files are correctly saved and able to loaded again? Thank you very much.

[EmbodiedGPT]

ok Let me check it and when I find the reason, I will update all the related files as soon as possible. I will review the entire section of the repo to ensure it operates correctly. This process will take some time, as some data and models are quite large and will require a while to upload. I will update as soon as possible.

[EmbodiedGPT]

When it is ready, I will tell you as soon as possible. Thanks for your great effort and support to this project.

[HaFred]

Another thing is about the video file as well. The z01 file here seems to be unable to unzip, isn't it supposed to be at least a zip file here to unzip?

• https://en.bandisoft.com/bandizip/howto/open-z01/

Yes, I totally get that open-sourcing the code is tedious. Thank you very much for your efforts!

[wofmanaf]

Another thing is about the video file as well. The z01 file here seems to be unable to unzip, isn't it supposed to be at least a zip file here to unzip?

• https://en.bandisoft.com/bandizip/howto/open-z01/

Yes, I totally get that open-sourcing the code is tedious. Thank you very much for your efforts!

Some files uploaded failed, we will reupload them, there are a total of 52 files, about 526G. Please wait for half a day or so until the files are ready.