Unexpected Random Track Prediction

[nikeshdevkota]

After training the trackformer model, I tried to evaluate it using python src/track.py. But I have been getting random predictions in between. For example, in frame 61, the tracking result is suddenly random.

I have used a custom CNN model as a backbone for my experiment on a custom dataset that contains small target objects.

This is the configuration in the track.yaml file. Is there any way to remove this randomness ?