TransVG

This is the official implementation of TransVG: End-to-End Visual Grounding with Transformers. This paper has been accepted by ICCV 2021.

@article{deng2021transvg,
  title={TransVG: End-to-End Visual Grounding with Transformers},
  author={Deng, Jiajun and Yang, Zhengyuan and Chen, Tianlang and Zhou, Wengang and Li, Houqiang},
  journal={arXiv preprint arXiv:2104.08541},
  year={2021}

}

Installation

Clone this repository.

git clone https://github.com/djiajunustc/TransVG

Prepare for the running environment.

You can either use the docker image we provide, or follow the installation steps in ReSC.
```
docker pull djiajun1206/vg:pytorch1.5
```

Getting Started

Please refer to GETTING_STARGTED.md to learn how to prepare the datasets and pretrained checkpoints.

Model Zoo

The models with ResNet-50 backbone are available in [Gdrive]

RefCOCO			RefCOCO+			RefCOCOg			ReferItGame
val	testA	testB	val	testA	testB	g-val	u-val	u-test	val	test
80.5	83.2	75.2	66.4	70.5	57.7	66.4	67.9	67.4	71.6	69.3

Training and Evaluation

Training

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --batch_size 8 --lr_bert 0.00001 --aug_crop --aug_scale --aug_translate --backbone resnet50 --detr_model ./checkpoints/detr-r50-referit.pth --bert_enc_num 12 --detr_enc_num 6 --dataset referit --max_query_len 20 --output_dir outputs/referit_r50 --epochs 90 --lr_drop 60

We recommend to set --max_query_len 40 for RefCOCOg, and --max_query_len 20 for other datasets.

We recommend to set --epochs 180 (--lr_drop 120 acoordingly) for RefCOCO+, and --epochs 90 (--lr_drop 60 acoordingly) for other datasets.

Evaluation

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --nproc_per_node=8 --use_env eval.py --batch_size 32 --num_workers 4 --bert_enc_num 12 --detr_enc_num 6 --backbone resnet50 --dataset referit --max_query_len 20 --eval_set test --eval_model ./outputs/referit_r50/best_checkpoint.pth --output_dir ./outputs/referit_r50

Acknowledge