TEM_scores

[DoublePan-Oh]

What's the meaning of the action start end in the TEM_scores files?

Thank you for your reply.

[sparkstj]

Hi, these are the actionness, starting, and ending probabilities in the corresponding frame.

[DoublePan-Oh]

I have some questions about feature extraction. in the paper, the frame stride is set to 8 for I3D , which means you divided each video into non-overlapping snippets with consecutive 8 frames, and 8 frames is a chuck? For example, if a video has N frames, after the I3D model, the final feature dimension is (N//8, C) ? Looking forward to your reply!

…

------------------ 原始邮件 ------------------ 发件人: "MCG-NJU/RTD-Action" ***@***.***>; 发送时间: 2022年3月2日(星期三) 下午2:41 ***@***.***>; ***@***.******@***.***>; 主题: Re: [MCG-NJU/RTD-Action] TEM_scores (Issue #17) Hi, these are the actionness, starting, and ending probabilities in the corresponding frame. — Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you authored the thread.Message ID: ***@***.***>

[sparkstj]

Yes, but the final feature dimension is co-determined by the final avg3dpooling layer in I3D. The pooling kernel we used is (8,7,7) and the stride is (1,1,1), therefore the feature dimension should be (N//8-7, C).

[Rishitha1234]

How to implement this code??
Can you explain me in detail (step by step)??

[sparkstj]

You can check this repo for detailed code. It is the exact code we use to generate thumos features, with a mild modification on the avgpooling layer after Mixed_5c (kernel from (2,7,7) to (8,7,7)) .