TransVTSpotter: End-to-end Video Text Spotter with Transformer

Introduction

A Multilingual, Open World Video Text Dataset and End-to-end Video Text Spotter with Transformer

Link to our MOVText: A Large-Scale, Multilingual Open World Dataset for Video Text Spotting

Updates

• (11/05/2022) TransDETR, a better transformer-based video text spotting method has been launched.

• (08/04/2021) Refactoring the code.

• (10/20/2021) The complete code has been released .

ICDAR2015(video) Tracking challenge

Methods	MOTA	MOTP	IDF1	Mostly Matched	Partially Matched	Mostly Lost
TransVTSpotter	45.75	73.58	57.56	658	611	647

Notes

• The training time is on 8 NVIDIA V100 GPUs with batchsize 16.
• We use the models pre-trained on COCOTextV2.
• We do not release the recognition code due to the company's regulations.

Demo

Installation

The codebases are built on top of Deformable DETR and TransTrack.

Requirements

• Linux, CUDA>=9.2, GCC>=5.4
• Python>=3.7
• PyTorch ≥ 1.5 and torchvision that matches the PyTorch installation. You can install them together at pytorch.org to make sure of this
• OpenCV is optional and needed by demo and visualization

Steps

1. Install and build libs

git clone [email protected]:weijiawu/TransVTSpotter.git
cd TransVTSpotter
cd models/ops
python setup.py build install
cd ../..
pip install -r requirements.txt

2. Prepare datasets and annotations

COCOTextV2 dataset is available in COCOTextV2.

python3 track_tools/convert_COCOText_to_coco.py

ICDAR2015 dataset is available in icdar2015.

python3 track_tools/convert_ICDAR15video_to_coco.py

3. Pre-train on COCOTextV2

python3 -m torch.distributed.launch --nproc_per_node=8 --use_env main_track.py  --output_dir ./output/Pretrain_COCOTextV2 --dataset_file pretrain --coco_path ./Data/COCOTextV2 --batch_size 2  --with_box_refine --num_queries 500 --epochs 300 --lr_drop 100 --resume ./output/Pretrain_COCOTextV2/checkpoint.pth

python3 track_tools/Pretrain_model_to_mot.py

The pre-trained model is available Baidu Netdisk， password:59w8. Google Netdisk

And the MOTA 44% can be found here password:xnlw. Google Netdisk

4. Train TransVTSpotter

python3 -m torch.distributed.launch --nproc_per_node=8 --use_env main_track.py  --output_dir ./output/ICDAR15 --dataset_file text --coco_path ./Data/ICDAR2015_video --batch_size 2  --with_box_refine  --num_queries 300 --epochs 80 --lr_drop 40 --resume ./output/Pretrain_COCOTextV2/pretrain_coco.pth

5. Inference and Visualize TransVTSpotter

# Inference
python3 main_track.py  --output_dir ./output/ICDAR15 --dataset_file text --coco_path ./Data/ICDAR2015_video --batch_size 1 --resume ./output/ICDAR15/checkpoint.pth --eval --with_box_refine --num_queries 300 --track_thresh 0.3

# Visualize
python3 track_tools/Evaluation_ICDAR15_video/vis_tracking.py

License

TransVTSpotter is released under MIT License.

Citing

If you use TranVTSpotter in your research or wish to refer to the baseline results published here, please use the following BibTeX entries:

@article{wu2021opentext,
  title={A Bilingual, OpenWorld Video Text Dataset and End-to-end Video Text Spotter with Transformer},
  author={Weijia Wu, Debing Zhang, Yuanqiang Cai, Sibo Wang, Jiahong Li, Zhuang Li, Yejun Tang, Hong Zhou},
  journal={35th Conference on Neural Information Processing Systems (NeurIPS 2021) Track on Datasets and Benchmarks},
  year={2021}
}