Source codes of CenterTrack++ in 2021 ICME Workshop on Big Surveillance Data Processing and Analysis

Overview

MOT Tracked object bounding box association (CenterTrack++)

New association method based on CenterTrack. Two new branches (Tracked Size and IOU) are added onto the original CenterTrack tracker. The proposed method enables the computation of IOU distance matrix for more accurate object association compared to single displacement offset in the original CenterTrack.

Modification to CenterTrack method, image modified from CenterTrack

Abstract

The recent development of multi-object tracking (MOT) on point-based joint detection and tracking methods has attracted much research attention. CenterTrack tracking algorithm is one of such promising methods. It achieves state-of-the-art tracking performance using a simple detection model and single-frame spatial offsets to localize objects and predict their associations in a single network. However, this method still suffers from high identity switches due to the inferior association method. Only point displacement distance matrix is used to associate objects, which is not robust to deal with occlusion scenarios. To reduce the high number of identity switches and improve the tracking accuracy, more effective spatial information should be used in association. In this paper, we propose to incorporate a simple tracked object bounding box and overlapping prediction based on the current frame onto the CenterTrack algorithm. Specifically, we propose a Intersection over Union (IOU) distance cost matrix in the association step instead of point displacement distance. We evaluate our proposed tracker on the MOT17 test dataset, showing that our proposed method can reduce identity switches significantly by 22.6% and obtain a notable improvement of 1.5% in IDF1 compared to the original CenterTrack’s under the same tracklet lifetime.

Main Contributions

  • Proposed two branches (tracked box size and IOU)on top of the existing CenterTrack method for IOU distance metric computation in object association
  • Evaluation the proposed method on MOT17 dataset and obtain significant reduction in IDs and notable improvements in tracking accuracy score

Two new branches

The idea of the proposed method is to enhance the original displacement only association. Inspired by the IOU distance in SORT and IOU-Tracker, IOU distance can be used for more accurate object association across frames. IOU distance is calculated as 1 - IOU(bounding box of detected object in the previous frame and the predicted tracked object bounding box in the previous frame based on the current frame)

Tracked Object Size prediction

In order to obtain the IOU distance, the bounding box of the tracked object in the previous frame should be learnt. In this project, two methods were used to learn the tracked bounding box.

Tracking_wh: Directly learn the width and height of the tracked object bounding box in the previous frame.

Tracking_ltrb: Learn the offsets of the left, top, right and bottom of bounding box from the tracked object center in the previous frame.

The tracking_wh(left) and tracking_ltrb(right) approach illustration.

IOU prediction

To further suppress inaccurate association, the IOU value of the tracked object bounding box in adjacent frames is learnt to provide a threshold to filter unlikely associations. We would set the IOU distance to infinity if IOU distance > IOU.

Association Method

Main results

Comparison with other SOTA tracker on MOT17 test set

Note: S= Spatial features, A=appearance features

Tracker Association Features MOTA IDF1 IDs
TubeTK S 63 58.6 4137
CenterTrack S 67.8 64.7 3039
Ours S 68.1 66.2 2352
SST A 52.4 49.5 8431
CTrackerV1 S+A 66.6 57.4 5529
DEFT S+A 66.6 65.4 2823
FairMOT S+A 73.7 72.3 3303

Ablative studies on tracked size prediction method

Tracking_wh

Association Method IDF1 MOTA IDs FP(%) FN(%)
DIS 69.2 66.2 219 3.9 29.5
IOU 71.1 66.7 204 3.6 29.3
Combined 70.9 66.2 233 3.9 29.6
DIS→IOU 70 66.2 218 3.9 29.5
IOU→DIS 69.8 66.8 185 3.6 29.2

Tracking_ltrb

Association Method IDF1 MOTA IDs FP(%) FN(%)
DIS 69.2 66.2 219 3.9 29.5
IOU 72.4 66.7 191 3.8 29.2
Combined 70.8 66.5 236 3.8 29.3
DIS→IOU 70.5 66.6 202 3.8 29.2
IOU→DIS 71.4 66.7 166 3.8 29.2

Installation

Please refer to INSTALL.md for installation instructions.

Training and Evaluation

  • Download the crowdhuman pretrained model from xinyizhou/CenterTrack MODEL ZOO.md to models
  • prepare the data and convert it into COCO format refer to the original CenterTrack repo.
  • change the dataset root directory data_dir in opt.py
  • ablative studies for tracking_wh and tracking_ltrb approach respectively with five association method (IOU,DIS,Combined, IOU→DIS, DIS→IOU)
sh experiments/mot17val_tracking_wh.sh

sh experiments/mot17val_tracking_ltrb.sh

The trained model on MOT17val dataset using two approach are available in google drive, tracking_ltrb_70val.pth, tracking_wh_70val.pth.

  • Train on full mot17 training set and run model on the test set for evaluation
sh experiments/mot17full.sh

The trained models on full MOT17 dataset using ltrb approach is available in the google drive.

Demo comparison

Occlusion case

Original CenterTrack (left) vs CenterTrack++ (right)

Object exiting the frame

Original CenterTrack (left) vs CenterTrack++ (right)

Acknowledgement

A large part of the code is adapted from xingyizhou/CenterTrack, thanks for their wonderful inspiration.

Citation

If you find this paper and code useful in your research, please cite our papers.

@misc{yang2021multiobject,
      title={Multi-object Tracking with Tracked Object Bounding Box Association}, 
      author={Nanyang Yang and Yi Wang and Lap-Pui Chau},
      year={2021},
      eprint={2105.07901},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}
Owner
Nanyang Technological University Information Engineering and Media Student
Using Python to Play Cyberpunk 2077

CyberPython 2077 Using Python to Play Cyberpunk 2077 This repo will contain code from the Cyberpython 2077 video series on Youtube (youtube.

Harrison 118 Oct 18, 2022
A community run, 5-day PyTorch Deep Learning Bootcamp

Deep Learning Winter School, November 2107. Tel Aviv Deep Learning Bootcamp : http://deep-ml.com. About Tel-Aviv Deep Learning Bootcamp is an intensiv

Shlomo Kashani. 1.3k Sep 04, 2021
Pervasive Attention: 2D Convolutional Networks for Sequence-to-Sequence Prediction

This is a fork of Fairseq(-py) with implementations of the following models: Pervasive Attention - 2D Convolutional Neural Networks for Sequence-to-Se

Maha 490 Dec 15, 2022
A TensorFlow implementation of the Mnemonic Descent Method.

MDM A Tensorflow implementation of the Mnemonic Descent Method. Mnemonic Descent Method: A recurrent process applied for end-to-end face alignment G.

123 Oct 07, 2022
Implements Stacked-RNN in numpy and torch with manual forward and backward functions

Recurrent Neural Networks Implements simple recurrent network and a stacked recurrent network in numpy and torch respectively. Both flavours implement

Vishal R 1 Nov 16, 2021
Python package to generate image embeddings with CLIP without PyTorch/TensorFlow

imgbeddings A Python package to generate embedding vectors from images, using OpenAI's robust CLIP model via Hugging Face transformers. These image em

Max Woolf 81 Jan 04, 2023
Malmo Collaborative AI Challenge - Team Pig Catcher

The Malmo Collaborative AI Challenge - Team Pig Catcher Approach The challenge involves 2 agents who can either cooperate or defect. The optimal polic

Kai Arulkumaran 66 Jun 29, 2022
A library for finding knowledge neurons in pretrained transformer models.

knowledge-neurons An open source repository replicating the 2021 paper Knowledge Neurons in Pretrained Transformers by Dai et al., and extending the t

EleutherAI 96 Dec 21, 2022
dataset for ECCV 2020 "Motion Capture from Internet Videos"

Motion Capture from Internet Videos Motion Capture from Internet Videos Junting Dong*, Qing Shuai*, Yuanqing Zhang, Xian Liu, Xiaowei Zhou, Hujun Bao

ZJU3DV 98 Dec 07, 2022
Implementation of ICLR 2020 paper "Revisiting Self-Training for Neural Sequence Generation"

Self-Training for Neural Sequence Generation This repo includes instructions for running noisy self-training algorithms from the following paper: Revi

Junxian He 45 Dec 31, 2022
Residual Dense Net De-Interlace Filter (RDNDIF)

Residual Dense Net De-Interlace Filter (RDNDIF) Work in progress deep de-interlacer filter. It is based on the architecture proposed by Bernasconi et

Louis 7 Feb 15, 2022
Collection of generative models in Tensorflow

tensorflow-generative-model-collections Tensorflow implementation of various GANs and VAEs. Related Repositories Pytorch version Pytorch version of th

3.8k Dec 30, 2022
Sequence to Sequence Models with PyTorch

Sequence to Sequence models with PyTorch This repository contains implementations of Sequence to Sequence (Seq2Seq) models in PyTorch At present it ha

Sandeep Subramanian 708 Dec 19, 2022
Awesome Long-Tailed Learning

Awesome Long-Tailed Learning This repo pays specially attention to the long-tailed distribution, where labels follow a long-tailed or power-law distri

Stomach_ache 284 Jan 06, 2023
Simple converter for deploying Stable-Baselines3 model to TFLite and/or Coral

Running SB3 developed agents on TFLite or Coral Introduction I've been using Stable-Baselines3 to train agents against some custom Gyms, some of which

Gary Briggs 16 Oct 11, 2022
Cancer-and-Tumor-Detection-Using-Inception-model - In this repo i am gonna show you how i did cancer/tumor detection in lungs using deep neural networks, specifically here the Inception model by google.

Cancer-and-Tumor-Detection-Using-Inception-model In this repo i am gonna show you how i did cancer/tumor detection in lungs using deep neural networks

Deepak Nandwani 1 Jan 01, 2022
Viewmaker Networks: Learning Views for Unsupervised Representation Learning

Viewmaker Networks: Learning Views for Unsupervised Representation Learning Alex Tamkin, Mike Wu, and Noah Goodman Paper link: https://arxiv.org/abs/2

Alex Tamkin 31 Dec 01, 2022
Release of SPLASH: Dataset for semantic parse correction with natural language feedback in the context of text-to-SQL parsing

SPLASH: Semantic Parsing with Language Assistance from Humans SPLASH is dataset for the task of semantic parse correction with natural language feedba

Microsoft Research - Language and Information Technologies (MSR LIT) 35 Oct 31, 2022
A project studying the influence of communication in multi-objective normal-form games

Communication in Multi-Objective Normal-Form Games This repo consists of five different types of agents that we have used in our study of communicatio

Willem Röpke 0 Dec 17, 2021
Compositional Sketch Search

Compositional Sketch Search Official repository for ICIP 2021 Paper: Compositional Sketch Search Requirements Install and activate conda environment c

Alexander Black 8 Sep 06, 2021