Issues after training VeRI dataset

[hghimanshu]

Hey @nwojke, I have trained the model on VeRI dataset for about 122k iterations. The (classification accuracy) value shown on the tensorboard at that time was about 0.98 so I stopped the training. Then I export the model to a .pb file. I am facing some issues regarding the deep sort implementation for vehicle tracking.

I am having a initial detection file (.txt file) which I get from my vehicle detector Below are some of my issues :-

• do I need to generate a .npy file for every video on which I am running my code?
• also, the .npy file mostly starts generating detection after the first 3 frames, how can I resolve this issue ?

[a0981456759]

hello, Could you share your model, thx

[hghimanshu]

well, this issue is solved.

[phungthengoc]

please, how to train cosine metric learning for VeriDataset