11-03-2024
: C++ implementation of our tracker is available now
10-11-2023
: ONNX-Runtime and TensorRT-based inference code is released. Now, our MVT runs at ~70 fps on CPU and ~300 fps on GPU :zap::zap:. Check the page for details.
14-09-2023
: The pretrained tracker model is released
13-09-2023
: The paper is available on arXiv now
22-08-2023
: The MVT tracker training and inference code is released
21-08-2023
: The paper is accepted at BMVC2023
Install the dependency packages using the environment file mvt_pyenv.yml
.
Generate the relevant files:
python tracking/create_default_local_file.py --workspace_dir . --data_dir ./data --save_dir ./output
After running this command, modify the datasets paths by editing these files
lib/train/admin/local.py # paths about training
lib/test/evaluation/local.py # paths about testing
lib/train/admin/local.py
pretrained_models/
folder63, 67, and 71
in the base_backbone.py file. Replace these lines with self.z_dict1 = template.tensors
.python tracking/train.py --script mobilevit_track --config mobilevit_256_128x1_got10k_ep100_cosine_annealing --save_dir ./output --mode single
output/logs/
folderThe pretrained tracker model can be found here
lib/test/evaluation/local.py
output/checkpoints/
folder python tracking/test.py --tracker_name mobilevit_track --tracker_param mobilevit_256_128x1_got10k_ep100_cosine_annealing --dataset got10k_test/trackingnet/lasot
DEVICE
variable between cuda
and cpu
in the --tracker_param
file for GPU and CPU-based inference, respectively output/test/
folderpython tracking/profile_model.py
MobileViT
backbone from ml-cvnets. Thank you!If our work is useful for your research, please consider citing:
@inproceedings{Gopal_2023_BMVC,
author = {Goutam Yelluru Gopal and Maria Amer},
title = {Mobile Vision Transformer-based Visual Object Tracking},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year = {2023},
url = {https://papers.bmvc2023.org/0800.pdf}
}