Could I share your extracted features for THUMOS 2014?

[YanZhang-bit]

Hi @piergiaj
Thanks for sharing your code!I have repeat your THUMOS'14 experiments in your paper "Temporal gaussian mixture layer for videos"(Table 6),but i can't achieve the performance mentioned in your paper.I think the issues maybe appears in the feature extraction。So Could you share your extracted features?Thank you very much!

[River-mao]

Hi @piergiaj
Thanks for sharing your code!I have repeat your THUMOS'14 experiments in your paper "Temporal gaussian mixture layer for videos"(Table 6),but i can't achieve the performance mentioned in your paper.I think the issues maybe appears in the feature extraction。So Could you share your extracted features?Thank you very much!

Hi! Have you solved this problem?

[ptirupat]

Hello @piergiaj

I used your code to extract I3D features (RGB only) from Multi-Thumos dataset and trained the TGM model with 3 layer, L = 5. The result I get is mAP of 0.330, which is less than the results (0.372) reported in Table 8 of your paper. I believe it is to do with feature extraction using I3D. Can you please share the code or the extracted features from this dataset? Thank you very much.