model mismatch

[upgirlnana]

i train the model with your code on combined dataset,when i try to test on single dataset,it occurs size mismatch bug

[xfanplus]

i train the model with your code on combined dataset,when i try to test on single dataset,it occurs size mismatch bug

I think the reason is the number of classes, you should change it accordingly.

[upgirlnana]

thank you very much. i test single dataset with the trained combined model and change the class number to 2220,and the test result of map and cmc1 are low on both holistic and partial dataset(market1501:cmc1: 84.83%,cuhk03:cmc1: 55.43%,duke:cmc1: 73.11%,partial_reid map:45%,Partial_iLIDS:cmc1: 36.13%]). i wonder if there is something wrong on parameter when i train,can you help me analyses this problem

[upgirlnana]

i train the model with your code on combined dataset,when i try to test on single dataset,it occurs size mismatch bug

I think the reason is the number of classes, you should change it accordingly. thank you very much. i test single dataset with the trained combined model and change the class number to 2220,and the test result of map and cmc1 are low on both holistic and partial dataset(market1501:cmc1: 84.83%,cuhk03:cmc1: 55.43%,duke:cmc1: 73.11%,partial_reid map:45%,Partial_iLIDS:cmc1: 36.13%]). i wonder if there is something wrong on parameter when i train,can you help me analyses this problem

[upgirlnana]

the result of single train single test can almost get the accuracy as the paper said. the problem occurs on combined model .Looking forward to your prompt reply.

[xfanplus]

the result of single train single test can almost get the accuracy as the paper said. the problem occurs on combined model .Looking forward to your prompt reply.

I don't test much on combined training in this implementation. If you wish to get better accuracies, you can add CUHK-SYSU training data, which will give you a higher score.

[upgirlnana]

if you can share your CUHK-SYSU training data.thank you very much

[Adorablepet]

@upgirlnana Have you solved about CUHK-SYSU training data?

[Dtappledoghuati]

When I run test and train, the following situations will occur. I am very helpless. May I know if you have encountered them? (scp) misstian@misstian-N9x0TD-TF:/media/misstian/tnq/Open-SCPNet-master$ python script/experiment/train.py -d '(0,)' --only_test false --dataset combined --last_conv_stride 1 --normalize_feature false --trainset_part trainval --exp_dir train --steps_per_log 10 --epochs_per_val 5 ------------------------------------------------------------Traceback (most recent call last):