search

Ning-Ding / Implementation-CVPR2015-CNN-for-ReID

Implementation for CVPR 2015 Paper: "An Improved Deep Learning Architecture for Person Re-Identification".
MIT License
147 stars 71 forks source link

how to compute CMC and rank-m ?? #29

Open Phoebe-star opened 6 years ago

Phoebe-star commented 6 years ago

how to compute CMC and rank-m ?? can you tell me thanks

bmiftah commented 6 years ago

I might be late to put a follow up issues on what @Phoebe-star As for the cmc and rank issues, I saw the part of the code dealing with cmc and rank issues inside the main file , but i didnt know if we have to write these output to a file and plot them later , Anybody tried listing the rank or writing them to a file and try to plot it ... I appreciate ...

hs-ucsb commented 6 years ago

@bmiftah Hi, I ran the main.py script. I think it will train the model. Can you please tell me how to use this model to predict test input?

bmiftah commented 6 years ago

To my understanding , the author already split test data and make prediction with this after the model was trained ... you see this in the main file with the call to cmc(....) , In side cmc(..) with the call to similarity_batch = model.predict_on_batch([camera_batch1, camera2]),the splited data is fetched with a _get_test_data() function, .. similarity prediction of probe against gallery and subsequent ranking based on these prediction is done in cmc( ..) ...

hs-ucsb commented 6 years ago

@bmiftah Hi, In the main method, there are parameters : train_num=100, one_epoch=30000, epoch_num=1, flag_random=None, random_pattern=lambda x:x/2+0.4, flag_train=0, flag_val=1, which_val_data='validation', nb_val_samples=1000

If possible, can you please explain what each parameters means? Especially,. whats the difference between one_epoch and train_num? Also, what does nb_val_samples mean ?