search

blue-blue272 / VideoReID-TCLNet

Code of Temporal Complementary Learning for Video Person Re-Identification (ECCV2020)
67 stars 19 forks source link

when I run the **ilidsvid** dateset,I got a similar problem.My best result is different from your papers. #8

Closed mogeko2333 closed 4 years ago

mogeko2333 commented 4 years ago

**Extracted features for query set, obtained torch.Size([150, 4096]) matrix Extracted features for gallery set, obtained torch.Size([150, 4096]) matrix Extracting features complete in 0m 3s Computing distance matrix Computing CMC and mAP Results ---------- mAP: 77.56% CMC curve Rank-1 : 69.33% Rank-5 : 88.67% Rank-10 : 92.00% Rank-20 : 96.67%

==> Best Rank-1 72.7%, achieved at epoch 100 Finished. Total elapsed time (h:m:s): 0:28:30. Training time (h:m:s): 0:27:31.**

It may be that the training result only used 4 frames. Can you tell me how to use all the frames ? Sorry for any possible interruption.

blue-blue272 commented 4 years ago

The test is "python test.py --root your path to ilidsvid dataset". And the dataset to load ilidsvid can be seen https://github.com/guxinqian/AP3D/blob/master/tools/data_manager.py

mogeko2333 commented 4 years ago

Thank you！I will try.