TrackFormer

This repository provides the official implementation of the TrackFormer: Multi-Object Tracking with Transformers paper by Tim Meinhardt, Alexander Kirillov, Laura Leal-Taixe and Christoph Feichtenhofer. The codebase builds upon DETR, Deformable DETR and Tracktor.

Abstract

The challenging task of multi-object tracking (MOT) requires simultaneous reasoning about track initialization, identity, and spatiotemporal trajectories.
We formulate this task as a frame-to-frame set prediction problem and introduce TrackFormer, an end-to-end MOT approach based on an encoder-decoder Transformer architecture.
Our model achieves data association between frames via attention by evolving a set of track predictions through a video sequence.
The Transformer decoder initializes new tracks from static object queries and autoregressively follows existing tracks in space and time with the new concept of identity preserving track queries.
Both decoder query types benefit from self- and encoder-decoder attention on global frame-level features, thereby omitting any additional graph optimization and matching or modeling of motion and appearance.
TrackFormer represents a new tracking-by-attention paradigm and yields state-of-the-art performance on the task of multi-object tracking (MOT17) and segmentation (MOTS20).

MOT17-03-SDP

MOTS20-07

Installation

We refer to our docs/INSTALL.md for detailed installation instructions.

Train TrackFormer

We refer to our docs/TRAIN.md for detailed training instructions.

Evaluate TrackFormer

In order to evaluate TrackFormer on a multi-object tracking dataset, we provide the src/track.py script which supports several datasets and splits interchangle via the dataset_name argument (See src/datasets/tracking/factory.py for an overview of all datasets.) The default tracking configuration is specified in cfgs/track.yaml. To facilitate the reproducibility of our results, we provide evaluation metrics for both the train and test set.

MOT17

Private detections

python src/track.py with reid
MOT17 MOTA IDF1 MT ML FP FN ID SW.
Train 68.1 67.6 816 207 33549 71937 1935
Test 65.0 63.9 1074 324 70443 123552 3528

Public detections (DPM, FRCNN, SDP)

python src/track.py with \
    reid \
    public_detections=min_iou_0_5 \
    obj_detect_checkpoint_file=models/mots20_train_masks/checkpoint.pth
MOT17 MOTA IDF1 MT ML FP FN ID SW.
Train 67.2 66.9 663 294 14640 94122 1866
Test 62.5 60.7 702 632 32828 174921 3917

MOTS20

python src/track.py with \
    dataset_name=MOTS20-ALL \
    obj_detect_checkpoint_file=models/mots20_train_masks/checkpoint.pth

Our tracking script only applies MOT17 metrics evaluation but outputs MOTS20 mask prediction files. To evaluate these download the official MOTChallengeEvalKit.

MOTS20 sMOTSA IDF1 FP FN IDs
Train -- -- -- -- --
Test 54.9 63.6 2233 7195 278

Demo

To facilitate the application of TrackFormer, we provide a demo interface which allows for a quick processing of a given video sequence.

ffmpeg -i data/snakeboard/snakeboard.mp4 -vf fps=30 data/snakeboard/%06d.png

python src/track.py with \
    dataset_name=DEMO \
    data_root_dir=data/snakeboard \
    output_dir=data/snakeboard \
    write_images=pretty

method

Publication

If you use this software in your research, please cite our publication:

@InProceedings{meinhardt2021trackformer,
    title={TrackFormer: Multi-Object Tracking with Transformers},
    author={Tim Meinhardt and Alexander Kirillov and Laura Leal-Taixe and Christoph Feichtenhofer},
    year={2021},
    eprint={2101.02702},
    archivePrefix={arXiv},
}

GitHub

https://github.com/timmeinhardt/trackformer