search

wangxiang1230 / OadTR

Code for our ICCV 2021 Paper "OadTR: Online Action Detection with Transformers".
MIT License
87 stars 12 forks source link

THUMOS14-Kinetics feature FPS. #10

Closed sqiangcao99 closed 2 years ago

sqiangcao99 commented 2 years ago

THUMOS14-Kinetics feature is extracted at 4 FPS in IDN. Is your experimental configuration consistent with IDN? In addition, can you provide more details about the feature extraction method?

wangxiang1230 commented 2 years ago

THUMOS14-Kinetics feature is extracted at 4 FPS in IDN. Is your experimental configuration consistent with IDN? In addition, can you provide more details about the feature extraction method?

Hi, thanks for your attention to OadTR. We extract feature in its original FPS. And the feature extraction method is here.

SCZwangxiao commented 2 years ago

THUMOS14-Kinetics feature is extracted at 4 FPS in IDN. Is your experimental configuration consistent with IDN? In addition, can you provide more details about the feature extraction method?

Hi, thanks for your attention to OadTR. We extract feature in its original FPS. And the feature extraction method is here.

Could you explain what does 'it' refer to here？If it refers to the video itself, your setting is different from IDN？

wangxiang1230 commented 2 years ago

THUMOS14-Kinetics feature is extracted at 4 FPS in IDN. Is your experimental configuration consistent with IDN? In addition, can you provide more details about the feature extraction method?

Hi, thanks for your attention to OadTR. We extract feature in its original FPS. And the feature extraction method is here.

Could you explain what does 'it' refer to here？If it refers to the video itself, your setting is different from IDN？

Thanks for your attention to our OadTR. 4FPS in IDN refers to getting 4 TSN features from one second of video, which may be misleading. Because in fact, each TSN feature is obtained from 6 frames of input, 4x6=24, that is, 24FPS. The original video frame rate is about 24. We extract video frames according to the original frame rate, and get a feature every 6 frames, which is consistent with IDN.