Suggestions to resolve ID swapping issue in two mice tracking videos #2457
Description
I am training sleap to track two mice in a box and I have about 20+ videos that I run inference on. The body tracking is working fine but the mouse ID tracking seems to suffer from constant swapping. This is espcially common occurence when mice are present in close proximity to one another but also occurs even when mice are far apart. I would like to fix this issue and improve my model and overall prediction output.
I am using multi-animal top-down-id pipeline and I have over 350+ labelled frames across my 20 videos.
Here are the exact settings for you to look:
Here are my current sleap-nn-track command settings that I use to run inference:
sleap-nn-track \ --data_path "$LABEL_FILE" \ --video_index "$IDX" \ --model_paths "$MODEL1" \ --model_paths "$MODEL2" \ -o "$OUTPUT_DIR/prediction_${IDX}.predictions.slp" \ --batch_size 10 \ --max_instances 2 \ --tracking \ --track_matching_method hungarian \ --tracking_window_size 60 \ --candidates_method local_queues \ --max_tracks 2 \ --tracking_target_instance_count 2 \ --use_flow \ --post_connect_single_breaks \ --scoring_reduction robust_quantile \ --features keypoints \ --scoring_method oks
I am also attaching screenshota as an example to demonstrate the common occurence where mice swap IDs between one frame to the next.
I would really appreciate any feedback to help me improve my sleap-tracking analysis.
Thank you for your continued support and hard work on the SLEAP package!