search

KPeng9510 / RelaMiX

MIT License
19 stars 0 forks source link

Issues about the dataset details and extracting I3D features #1

Closed wyyy04 closed 7 months ago

wyyy04 commented 9 months ago

Thanks for your excellent work! However, I find some difficulties in reproducing the code. I look forward to your reply and answer！

  1. ToyotSmartHome dataset I would like to ask if the ToyotSmartHome dataset used in the article is a trimmed or untrimmed version. Whether the official I3D features are used？

  2. Extracting I3D features It is pointed out in the paper and instructions that the frame-level I3D features of each video need to be extracted before training and testing the network, but I did not find the relevant code in the repository. Instead, the code in [dataset_preparation] uses the C3D network and ResNet to extract video frame features. Could you please provide the code for extracting features with I3D?

  3. calculate_statistics_trainset_X.py Several functions in these python files are not defined. For example, in calculate_statistics_trainset_hmdb_ucf.py Line 257: prec1 = validate(val_loader, model, criterion, num_class, epoch, val_file) Line 264: val_short_file.write('%.3f\n' % prec1) Line 272: val_best_file.write('%.3f\n' % best_prec1) validate/val_short_file/val_best_file are not defined. Please check it.

These are my issues, and I'm looking forward to your reply. Thanks for your help!

KPeng9510 commented 9 months ago

Hi,

Thank you for your interests on our work.

  1. Our work is based on the trimmed ToyotaSmartHome. We do not use the official features.

  2. Please refer to this repo for the feature extraction, we follow the same way with theirs. https://github.com/ldkong1205/TranSVAE

  3. you could delete those lines which are not useful, we only use the stored pkl files.