search

guxinqian / AP3D

Pytorch implementation of "Appearance-Preserving 3D Convolution for Video-based Person Re-identification"
Apache License 2.0
97 stars 24 forks source link

Replicate only I3D results #5

Closed abhishekaich27 closed 4 years ago

abhishekaich27 commented 4 years ago

Hi Xinqian,

To replicate only I3D results, do we need to change anything else except AP3D.APP3DC ---> AP3D.I3D in L168 of ResNet.py? If I only do this change, I get 3% mAP difference according to Table 1.

guxinqian commented 4 years ago

Yes, only L168 needs to be changed. Batchsize, the num of GPUs and the version of pytorch may impact the results. In general, 1% error is reasonable. In addition, running test-all.py is necessary, because the process of test in train.py only uses 4 frames per video.