PyTorch implementation of HDN(Homography Decomposition Networks) for planar object tracking

Last update: Dec 15, 2022

Related tags

Overview

Homography Decomposition Networks for Planar Object Tracking

This project is the offical PyTorch implementation of HDN(Homography Decomposition Networks) for planar object tracking. (AAAI 2022, Accepted)

Project Page | Paper

@misc{zhan2021homography,
      title={Homography Decomposition Networks for Planar Object Tracking}, 
      author={Xinrui Zhan and Yueran Liu and Jianke Zhu and Yang Li},
      year={2021},
      eprint={2112.07909},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Installation

Please find installation instructions in INSTALL.md.

Quick Start: Using HDN

Add HDN to your PYTHONPATH

vim ~/.bashrc
# add home of project to PYTHONPATH
export PYTHONPATH=/path/to/HDN:/path/to/HDN/homo_estimator/Deep_homography/Oneline_DLTv1:$PYTHONPATH

Download models

Google Drive or Baidu Netdisk (key: 8uhq)

Base Setting

The global parameters setting file is hdn/core/config.py You first need to set the base path:

__C.BASE.PROJ_PATH = /xxx/xxx/project_root/ #/home/Kay/SOT/server_86/HDN/   (path_to_hdn)
__C.BASE.BASE_PATH = /xxx/xxx/ #/home/Kay/SOT/                  (base_path_to_workspace)
__C.BASE.DATA_PATH = /xxx/xxx/data/POT  #/home/Kay/data/POT     (path to POT datasets)
__C.BASE.DATA_ROOT = /xxx/xxx   #/home/Kay/Data/Dataset/        (path to other datasets)

Demo

Planar Object Tracking and its applications we provide 4 modes:

tracking: tracking planar object with not less than 4 points in the object.
img_replace: replacing planar object with image .
video_replace: replacing planar object with video.
mosiac: adding mosiac to planar object.

python tools/demo.py 
--snapshot model/hdn-simi-sup-hm-unsup.pth 
--config experiments/tracker_homo_config/proj_e2e_GOT_unconstrained_v2.yaml 
--video demo/door.mp4 
--mode img_replace 
--img_insert demo/coke2.jpg #required in mode 'img_replace'  
--video_insert demo/t5_videos/replace-video/   #required in mode 'video_replace'
--save # whether save the results.

e.g.

python tools/demo.py  --snapshot model/hdn-simi-sup-hm-unsup.pth  --config experiments/tracker_homo_config/proj_e2e_GOT_unconstrained_v2.yaml --video demo/door.mp4 --mode img_replace --img_insert demo/coke2.jpg --save

we provide some real-world videos here

Download testing datasets

POT

For POT dataset, download the videos from POT280 and annotations from here

1. unzip POT_v.zip and POT_annotation.zip and put them in your cfg.BASE.DATA_PATH #unzip the zip files
  cd POT_v
  unzip "*.zip"
  cd ..

2. mkdir POT
   mkdir path_to_hdn/testing_dataset
   python path_to_hdn/toolkit/benchmarks/POT/pot_video_to_pic.py #video to images  
   ln -s path_to_data/POT  path_to_hdn/testing_dataset/POT #link to testing_datasets


4. python path_to_hdn/toolkit/benchmarks/POT/generate_json_for_POT.py --dataset POT210 #generate json annotation for POT
   python path_to_hdn/toolkit/benchmarks/POT/generate_json_for_POT.py --dataset POT280

UCSB & POIC

Download from here put them in your cfg.BASE.DATA_PATH

ln -s path_to_data/UCSB  path_to_hdn/testing_dataset/UCSB #link to testing_datasets

generate json:

  python path_to_hdn/toolkit/benchmarks/POIC/generate_json_for_poic.py #generate json annotation for POT
  python path_to_hdn/toolkit/benchmarks/UCSB/generate_json_for_ucsb.py #generate json annotation for POT

Other datsets:

Download datasets and put them into testing_dataset directory. Jsons of commonly used datasets can be downloaded from here. If you want to test tracker on new dataset, please refer to pysot-toolkit to setting testing_dataset.

Test tracker

test POT

cd experiments/tracker_homo_config
python -u ../../tools/test.py \
	--snapshot ../../model/hdn-simi-sup-hm-unsup.pth \ # model path 
	--dataset POT210 \ # dataset name
	--config proj_e2e_GOT_unconstrained_v2.yaml # config file
	--vis   #display video

The testing results will in the current directory(./results/dataset/model_name/)

Eval tracker

For POT evaluation

1.use tools/change_pot_results_name.py to convert result_name(you need to set the path in the file).

2.use tools/convert2Homography.py to generate the homo file(you need to set the corresponding path in the file).

3.use POT toolkit to test the results. My version toolkit can be found here or official for other trackers:

For others:

For POIC, UCSB or POT evaluation on centroid precision, success rate, and robustness etc. assuming still in experiments/tracker_homo_config

python ../../tools/eval.py 	 \
	--tracker_path ./results \ # result path
	--dataset POIC        \ # dataset name
	--num 1 		 \ # number thread to eval
	--tracker_prefix 'model'   # tracker_name

The raw results can be downloaded at Google Drive or Baidu Netdisk (key:d98h)

Training 🔧

We use the COCO14 and GOT10K as our traning datasets. See TRAIN.md for detailed instruction.

Acknowledgement

This work is supported by the National Natural Science Foundation of China under Grants (61831015 and 62102152) and sponsored by CAAI-Huawei MindSpore Open Fund.

Our codes is based on SiamBAN and DeepHomography.

License

This project is released under the Apache 2.0 license.

PyTorch implementation of HDN(Homography Decomposition Networks) for planar object tracking

Related tags

Overview

Homography Decomposition Networks for Planar Object Tracking

Project Page | Paper

Installation

Quick Start: Using HDN

Add HDN to your PYTHONPATH

Download models

Base Setting

Demo

Download testing datasets

Test tracker

Eval tracker

For POT evaluation

For others:

Training 🔧

Acknowledgement

License

Owner

CaptainHook

implicit displacement field

Image augmentation library in Python for machine learning.

Semi-supevised Semantic Segmentation with High- and Low-level Consistency

Reference implementation of code generation projects from Facebook AI Research. General toolkit to apply machine learning to code, from dataset creation to model training and evaluation. Comes with pretrained models.

SAGE: Sensitivity-guided Adaptive Learning Rate for Transformers

A Library for Modelling Probabilistic Hierarchical Graphical Models in PyTorch

The code repository for "RCNet: Reverse Feature Pyramid and Cross-scale Shift Network for Object Detection" (ACM MM'21)

Nonnegative spatial factorization for multivariate count data

GraphGT: Machine Learning Datasets for Graph Generation and Transformation

Adversarial Learning for Modeling Human Motion

Live training loss plot in Jupyter Notebook for Keras, PyTorch and others

Tightness-aware Evaluation Protocol for Scene Text Detection

An Unsupervised Detection Framework for Chinese Jargons in the Darknet

Task-related Saliency Network For Few-shot learning

AI Toolkit for Healthcare Imaging

Official code repository of the paper Learning Associative Inference Using Fast Weight Memory by Schlag et al.

Federated Learning Based on Dynamic Regularization

PyTorch implementation of Lip to Speech Synthesis with Visual Context Attentional GAN (NeurIPS2021)

StudioGAN is a Pytorch library providing implementations of representative Generative Adversarial Networks (GANs) for conditional/unconditional image generation.

This GitHub repository contains code used for plots in NeurIPS 2021 paper 'Stochastic Multi-Armed Bandits with Control Variates.'