search

Tencent / ActionDetection-DBG

Code for AAAI2020 paper "Fast Learning of Temporal Action Proposal via Dense Boundary Generator"
Other
347 stars 105 forks source link

Is that possible to release the detection demo? #15

Open semchan opened 4 years ago

semchan commented 4 years ago

Greeting! since there is only evaluation for proposals, do you plan to release the detection code? thanks.

semchan commented 4 years ago

HI, I use DBG to produce the proposals from the feature's data which you provided, and then use the proposals and the TSN features directly to do the temporal detection, the result is poor, only got 0.2mPA. I wonder, the big problem is I have not use the final video-level score for detection, but use only the 100*400 feature's for detection. Is that right? and would you please give me the guide to improve the mAP based on DBG? thanks very much.

[INIT] Loaded annotations from validation subset. Number of ground truth instances: 7654 Number of predictions: 472700 Fixed threshold for tiou score: [0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95] [RESULTS] Performance on ActivityNet detection task. Average-mAP: 0.20763225314731745

happyCodingSusan commented 4 years ago

Great work! Thank you for sharing. I also want to ask whether you could release your detection demo.

quanh1990 commented 4 years ago

@lijiannuist Would you please release the detection demo? Thanks a lot. I still cannot see how to get the action label and the start/end using your code for a single video.

GeLee-Q commented 3 years ago

请问inference model到时候会开源出来嘛，想试试能否实际的去应用一下。