Home

Awesome

Deep SORT

Introduction

This repository contains code for Simple Online and Realtime Tracking with a Deep Association Metric (Deep SORT). We extend the original SORT algorithm to integrate appearance information based on a deep appearance descriptor. See the arXiv preprint for more information.

Dependencies

The code is compatible with Python 2.7 and 3. The following dependencies are needed to run the tracker:

Additionally, feature generation requires TensorFlow (>= 1.0).

Installation

First, clone the repository:

git clone https://github.com/nwojke/deep_sort.git

Then, download pre-generated detections and the CNN checkpoint file from here.

NOTE: The candidate object locations of our pre-generated detections are taken from the following paper:

F. Yu, W. Li, Q. Li, Y. Liu, X. Shi, J. Yan. POI: Multiple Object Tracking with
High Performance Detection and Appearance Feature. In BMTT, SenseTime Group
Limited, 2016.

We have replaced the appearance descriptor with a custom deep convolutional neural network (see below).

Running the tracker

The following example starts the tracker on one of the MOT16 benchmark sequences. We assume resources have been extracted to the repository root directory and the MOT16 benchmark data is in ./MOT16:

python deep_sort_app.py \
    --sequence_dir=./MOT16/test/MOT16-06 \
    --detection_file=./resources/detections/MOT16_POI_test/MOT16-06.npy \
    --min_confidence=0.3 \
    --nn_budget=100 \
    --display=True

Check python deep_sort_app.py -h for an overview of available options. There are also scripts in the repository to visualize results, generate videos, and evaluate the MOT challenge benchmark.

Generating detections

Beside the main tracking application, this repository contains a script to generate features for person re-identification, suitable to compare the visual appearance of pedestrian bounding boxes using cosine similarity. The following example generates these features from standard MOT challenge detections. Again, we assume resources have been extracted to the repository root directory and MOT16 data is in ./MOT16:

python tools/generate_detections.py \
    --model=resources/networks/mars-small128.pb \
    --mot_dir=./MOT16/train \
    --output_dir=./resources/detections/MOT16_train

The model has been generated with TensorFlow 1.5. If you run into incompatibility, re-export the frozen inference graph to obtain a new mars-small128.pb that is compatible with your version:

python tools/freeze_model.py

The generate_detections.py stores for each sequence of the MOT16 dataset a separate binary file in NumPy native format. Each file contains an array of shape Nx138, where N is the number of detections in the corresponding MOT sequence. The first 10 columns of this array contain the raw MOT detection copied over from the input file. The remaining 128 columns store the appearance descriptor. The files generated by this command can be used as input for the deep_sort_app.py.

NOTE: If python tools/generate_detections.py raises a TensorFlow error, try passing an absolute path to the --model argument. This might help in some cases.

Training the model

To train the deep association metric model we used a novel cosine metric learning approach which is provided as a separate repository.

Highlevel overview of source files

In the top-level directory are executable scripts to execute, evaluate, and visualize the tracker. The main entry point is in deep_sort_app.py. This file runs the tracker on a MOTChallenge sequence.

In package deep_sort is the main tracking code:

The deep_sort_app.py expects detections in a custom format, stored in .npy files. These can be computed from MOTChallenge detections using generate_detections.py. We also provide pre-generated detections.

Citing DeepSORT

If you find this repo useful in your research, please consider citing the following papers:

@inproceedings{Wojke2017simple,
  title={Simple Online and Realtime Tracking with a Deep Association Metric},
  author={Wojke, Nicolai and Bewley, Alex and Paulus, Dietrich},
  booktitle={2017 IEEE International Conference on Image Processing (ICIP)},
  year={2017},
  pages={3645--3649},
  organization={IEEE},
  doi={10.1109/ICIP.2017.8296962}
}

@inproceedings{Wojke2018deep,
  title={Deep Cosine Metric Learning for Person Re-identification},
  author={Wojke, Nicolai and Bewley, Alex},
  booktitle={2018 IEEE Winter Conference on Applications of Computer Vision (WACV)},
  year={2018},
  pages={748--756},
  organization={IEEE},
  doi={10.1109/WACV.2018.00087}
}