Awesome

RGBD1K: A Large-Scale Dataset and Benchmark for RGB-D Object Tracking [AAAI2023]

The official implementation of the SPT tracker of the [AAAI2023] paper RGBD1K: A Large-Scale Dataset and Benchmark for RGB-D Object Tracking

Usage

Installation

Install the environment using Anaconda

conda create -n spt python=3.6
conda activate spt
bash install_pytorch17.sh

Data Preparation

The training dataset is the RGBD1K

--RGBD1K
    |--Adapter
        |--adapter1
        |--adapter2
        ...
    |--Animal
       |--alpaca1
       |--bear1
        ...
    ... 

Set project paths

Run the following command to set paths for this project

python tracking/create_default_local_file.py --workspace_dir . --data_dir ./data --save_dir .

After running this command, you can also modify paths by editing these two files

lib/train/admin/local.py  # paths about training
lib/test/evaluation/local.py  # paths about testing

Training

Download the pretrained Stark-s model and put it under ./pretrained_models/. Set the MODEL.PRETRAINED path in ./experiments/spt/rgbd1k.yaml

Training with multiple GPUs using DDP (4 RTX3090Ti with batch size of 16)

export PYTHONPATH=/path/to/SPT:$PYTHONPATH
python -m torch.distributed.launch --nproc_per_node=4 ./lib/train/run_training.py  

or using single GPU:

python ./lib/train/run_training.py  

Test

Edit ./lib/test/evaluation/local.py to set the test set path, then run

python ./tracking/test.py

You can also use the trained model for test.

Acknowledgment

• This repo is based on Stark which is an excellent work.

Contact

If you have any question, please feel free to contact us(xuefeng_zhu95@163.com)