search

Echo0125 / MAT-Memory-and-Anticipation-Transformer

[ICCV 2023] Official implementation of Memory-and-Anticipation Transformer for Online Action Understanding
Apache License 2.0
44 stars 3 forks source link

Request for the extracted features of TV Series and HDD datasets #12

Closed Prism-hua closed 5 months ago

Prism-hua commented 6 months ago

I have seen that the author cannot share the extracted features owing to the copyright. I would really appreciate it if anybody can share the pre-extracted features of TV Series and HDD with me or enlighten me with the detailed ways to extract the features myself.

Hello @mugenggeng, sorry to bother, I have read about the issue you opened in request for the features, and I am wondering if you have finished the extraction. If yes, I sincerelly hope that you can share that with me, Thanks a lot!

mugenggeng commented 6 months ago

As you said, due to copyright reasons, I can't share it with you. Because these two data sets are not public.

Prism-hua commented 6 months ago

Thanks @mugenggeng and hello @Echo0125, I am aware of the fact that these two datasets should not be shared through the public internet, I have tried to apply for these datasets myself. But as published in the readme file, ways to extract the features of HDD dataset and the Data Structure of HDD still remain unknown. Could you please share me with the procedure or any tutorial that I can follow to extract the features of HDD dataset and reproduce the results on these two datasets?

Echo0125 commented 6 months ago

Thanks for your interest in our work. For HDD, you will get the corresponding sensor features after applying for the dataset, you can just follow the structure in Data Structure and add the corresponding information to data/data_info.json.

Prism-hua commented 6 months ago

Thanks for your timely response @Echo0125. Embarrassed to tell you that I am also having problems extracting features of TVSeries. As the README of LSTR said, I should use MMAction2 and BN-Inception to get the RGB and optical flow inputs respectively, however, the urls of "checkpoint" and "here" does not work now, could please tell me what I should do to solve this? Thank you so much for your generous help!

Prism-hua commented 6 months ago

It seems that I have made a mistake, I should only use the checkpoints from MMAaction2, and use ResNet-50 and BN-Inception network to extract the rgb and optical flow inputs respectively. But now the problem is that nither do I have access to the checkpoints, nor do I know how to build feature extraction network using the two structures. Could you please provide me with more detailed infomation? Words fail to express my gratitude!!

Echo0125 commented 6 months ago

You can find the checkpoint in this branch mmaction2_0.x, just follow the file provided by mmaction2 to extract the feature: feature_extractor.