Inference

[LisaBernhardt]

I am trying to reproduce the inference as described in your repo.

I use single-gpu testing and run the following command in my terminal: python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} --out test.pkl --eval track config file: https://github.com/SysCV/qdtrack/blob/master/configs/bdd100k/qdtrack-frcnn_r50_fpn_12e_bdd100k.py checkpoint file: https://drive.google.com/file/d/1YNAQgd8rMqqEG-fRj3VWlO4G5kdwJbxz/view?usp=sharing

After running the command I load the .pkl file and visualize the first element of "['track_result]" on the first image of the validation dataset ( b22e02cd-6af68e18/b22e02cd-6af68e18-0000001.jpg). This is the output:

(TrackID_ClassID_predictionscore - what is written in top of each bounding box)

Then I use the same config and same checkpoint to load my model, and run the following function def simple_test(self, img, img_metas, rescale=False) (https://github.com/SysCV/qdtrack/blob/master/qdtrack/models/mot/qdtrack.py - line 83), with the same image and same image metas as above.

This is the output:

How is it possible that I get two different outputs, when I am doing the same thing in two different ways ? Thank you very much for your help!