search

megvii-research / MOTRv2

[CVPR2023] MOTRv2: Bootstrapping End-to-End Multi-Object Tracking by Pretrained Object Detectors
Other
357 stars 45 forks source link

Evaluation metrics #68

Open AlirezaKh1990 opened 4 months ago

AlirezaKh1990 commented 4 months ago

I have downloaded the trained model and have trained one myself. I used eval.sh but it doesn't generate the metrics. How can I get the performance metrics? Does anyone know?

Shawnnnnn commented 3 months ago

You should comment out this section of code in eval_dance.sh:

python3 submit_dance.py ${args} --resume checkpoint.pth --exp_name tracker

and uncomment:

python3 ../TrackEval/scripts/run_mot_challenge.py \
    --SPLIT_TO_EVAL val  \
    --METRICS HOTA CLEAR Identity  \
    --GT_FOLDER /data/datasets/DanceTrack/val \
    --SEQMAP_FILE seqmap \
    --SKIP_SPLIT_FOL True \
    --TRACKER_SUB_FOLDER tracker \
    --TRACKERS_TO_EVAL $OUTPUT_DIR \
    --USE_PARALLEL True \
    --NUM_PARALLEL_CORES 8 \
    --PLOT_CURVES False \
    --TRACKERS_FOLDER '' | tee -a $OUTPUT_DIR/eval.log

you should git clone TrackEval and put it right location then run eval_dance.sh