chenhao2345
chenhao2345
So adding the intra-camera loss only makes the model focus more on intra-camera matching.
I think so. Because my conrigurations are tuned on 4 Gpus, which means 8 images per GPU. This setting usually has some affects on results.
Thank you. Glad to know this.
Hi, loss_ccl corresponds to L_agnostic in Equation (5) of the paper. Our results are obtained batch size and learning rate tuned with **4 GPUs**. Please try to run our code...
The problem is related to clustering. It seems to me that you didn't load the pretrained weights for the identity encoder. Could you check if you saved weights from stage...
Please first check you are working on Pytorch 1.2. Some people said newer Pytorch version might raise a problem, because of the updated syntax.
Please first check whether you are working on Pytorch 1.2. Some people said newer Pytorch version might raise a problem, because of the updated syntax.
Hi, thanks for your attention. I've never met this problem. It seems that this problem is related to CUDA. I used 1 GPU under cuda10.0 and cudnn7.4.
I think there may be some syntax updates in Pytorch between version 1.2.0 and 1.6.0.
你好，感谢关注 1. 在stage2里，我们固定住了id encoder，只预热生成器的部分，这个阶段有关id的指标(rank1 mAP)不会变。stage3里会变。 2. 跨域只需要使用一个跨域的baseline算法去预热id encoder。例如下载我们分享的跨域训练权重，然后修改train_stage2_market.sh里面`--init ./examples/logs/JVTC/market/resnet50_duke2market_epoch00100.pth`