Joshua Z. Zhang

@yhl41001 You can set nms threshold to 0 to disable NMS, you will get the network forward time, that's the best you can get. Typically I would suggest using mxnet...

You can extract the forward code from the evaluater I wrote, disable nms by setting threshold to 0, and feed in fake data.

Ground-truths changing all the time after augmentation. So it's better to put that into layers.

these are passed by **kwargs

I think the trick should work, make sure you fix the seeds before importing any module in train.py. Let me know if it still fails.

Fine-tuning is simply calling train.py with --finetune, make sure you fix early layers if you training data is not large enough. Training jointly is better IMO, but you can definitely...

I would use training data as validation set and see if the MAP is still very low. If true, then there must be some problems with the data. Otherwise, it...

@matanhs I am not getting your point, do you mean the the pre-trained models downloaded getting 0 mAP?

Don't hide the nan problem, which means you have bad parameters set or something wrong with the data. See if you still get nan with smaller learning rate

That is validation metric, train metric was not affected