Houjing Huang

Nice explanation, now it makes sense to me that more queries means better averaging and leads to more stable evaluation results. And my quick and rough modification to the `Market1501`...

- You can try to increase number of identities to 32 (~9600MB GPU) or 64 (two 12GB GPUs). - Set the margin to 0.3 and then train for longer time,...

Maybe I did not understand your requirements. You want to reproduce the results of paper *In Defense of the Triplet Loss for Person Re-Identification*? If it's true, you have to...

Thanks again. Yeah, maybe it's necessary to make the mAP eval code self-contained in open-reid. Looking forward to your solution.

Yeah, currently it is also my awkward solution to downgrade the package and check the version in the code.

@yee-kevin 1. You should use 0.18.1 2. If you are using open-reid which recommends using Python 3, then you should use Python 3 to reduce potential inconsistency. 3. Sklearn version...

这个可以参考另一个工程中[这个文件](https://github.com/huanghoujing/person-reid-triplet-loss-baseline/blob/bca5c27832797cfa87ca33af5a97f168acd0dd51/script/experiment/visualize_rank_list.py)。基本流程就是 - 把测试集的特征都提好 - 计算query图片和gallery图片之间的距离 - 对一个query图片来说，把其gallery图片中与query**相同id**且**相同camera**的图片去掉，然后考虑剩下的gallery，对query-gallery距离进行从小到大排序 - 取排序在前k个的gallery图片进行显示

Hi, these `**.num_batches_tracked` are features of newer versions of pytorch. It does not have any effect in my case.

您好，感谢关注！ 性能提升来自于 - stride = 1 - 增加训练迭代次数，300 epochs - backbone后面不引入全连接层 另外，batch内P（多少人）、K（每个人多少张图片）两个参数也会影响性能

Hi, can you run it under pytorch 0.3 and python 2.7 and report the results? I haven't tested it on pytorch 0.4 and python 3.