It doesn't work when test on my own datasets

[sry948749099]

Does anyone complete the project and test on your own datasets?
Why are only two ID tracked when test on my own datasets? but there are seven persons in my datasets.
The ID switch is also very terrible，so , can someone tell me how to change the project?

[alexHxun]

I think this project only focus on the special data like the demo videos, because the resnet model used for REID maybe optimized with the demo videos . I found the distance between each embedding of different people is quiet different, so the cluster algorithm can work well. if you change test data, it will get some problems.