Code for DeepSORT based on detector from TrackR-CNN for the Multi Object Tracking and Segmentation (MOTS) task.
Tracking part author: Zhiye Wen, Yan Wang
https://www.vision.rwth-aachen.de/media/papers/mots-multi-object-tracking-and-segmentation/MOTS.pdf
https://arxiv.org/abs/1703.07402
Our codes are stored in the forwarding/tracking/tracking_deepsort, the main function deep_sort_app is imported to forwarding/tracking/TrackingForwarder.py to combine with other parts in Track R-CNN. The results of detection are stored in forwarded/conv3d_sep2/detection/5/ as the input of tracking.
You can use the following command to run the our tracking algorithm and to obtain final results in the forwarded/conv3d_sep2/tracking_data :
python main.py configs/conv3d_sep2 "{\"build_networks\":false,\"import_detections\":true,\"task\":\"forward_tracking\",\"dataset\":\"KITTI_segtrack_feed\",\"do_tracking\":true,\"visualize_detections\":false,\"visualize_tracks\":false,\"load_epoch_no\":5,\"video_tags_to_load\":[\"0002\",\"0006\",\"0007\",\"0008\",\"0010\",\"0013\",\"0014\",\"0016\",\"0018\"]}"
You can also visualize the tracking results here by setting visualize_tracks to true, and results will be stored in forwarded/conv3d_sep2/vis/.
Run the script for the evaluation on the validation set
To evaluate, run
python mots_eval/eval.py forwarded/conv3d_sep2/tracking_data gt_folder val.seqmap
where "val.seqmap" is a textfile containing the sequences which you want to evaluate on.
The script for random tuning will find the best combination of tracking parameters on the training set and then evaluate these parameters on the validation set.
To use this script, run
python segtrack_tune_experiment.py forwarded/conv3d_sep2/detections/5 /srv/store/dlenv/home/users/pp5-y7s/Tr4_mahal/gt/instances_txt / /srv/store/dlenv/home/users/pp5-y7s/Tr4_mahal/evalresult /srv/store/dlenv/home/users/pp5-y7s/Tr4_mahal/tmp_folder mots_eval/ reid num_iterations
where /forwarded/conv3d_sep2/detections/5/ is a folder containing the model output on the training set (obtained by the forwarding command above); /mots_eval// refers to the official evaluation script; reid is association_type; num_iterations is the number of random trials (1000 in the paper Track R-CNN); /gt/instances_txt/ refers to the instances or instances_txt folder containing the annotations (which you can download from the project website); at /evalresult, a file will be created containing the results of the individual tuning iterations, please make sure this path is writable; at /tmp_folder a lot of intermediate folders will be stored.
Parts of this code are based on Nwojke(https://github.com/nwojke/deep_sort/tree/master/deep_sort)