Crop-You-Want-by-Text

A tool that allows you to crop multiple desired areas of multiple images based on Grounding DINO.

Install

**Note:**git

If you have a CUDA environment, please make sure the environment variable CUDA_HOME is set. It will be compiled under CPU-only mode if no CUDA available.

Installation:

Clone this repository from GitHub.

git clone https://github.com/orange-36/Crop-You-Want-by-Text.git

Init submodule.

cd Crop-You-Want-by-Text/
git submodule init
git submodule update

Change the current directory to the GroundingDINO folder.

cd GroundingDINO/

Install the required dependencies in the current directory.

pip install -e .
cd ..

Download pre-trained model weights.

mkdir weights
cd weights
wget -q https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth
cd ..

Getting Started

Execute the program and give a text prompt to cut out the desired part of the image.

python crop_you_want.py --image_path <image_path> --text_prompt "<text_prompt>"

Parameter:

• --image_path The input image path, you can input multiple paths or the directory path containing images to be processed.
• --text_prompt Enter a text prompt, which can be a word or a phrase. Use . to separate different text categories.
• --box_threshold Threshold for bounding box. (default: 0.25)
• --text_threshold Threshold to judge whether it is the corresponding text category. (default: 0.25)
• --extend Extra dilated target box to crop. (default: 0)
• --model_config Model config used by GroundingDINO. (defalt: "groundingdino/config/GroundingDINO_SwinT_OGC.py")
• --model_weight Pretrained model weights used by GroundingDINO. (defalt: "weights/groundingdino_swint_ogc.pth")
• --output_path Where to save the results. (default: "output/")
• --device Device want to use. If no gpu is available, set "cpu". (default: "cuda")
• --output_order The order of the output results, set score to go from high to low according to the score, or set x or y to go from left to right or top to bottom. (default: score)
• --no_sub_dir Do not create additional subdirectories, take effect by entering --no_sub_dir.
• --square_crop Crop the image into a square, take effect by set --square_crop.

Execute the following program to obtain the example results.

python crop_you_want.py \
  --image_path images/man.png \
  --text_prompt "eyes . mouth .  ears . nose . eyebrows" \
  --box_threshold 0.3 \
  --text_threshold 0.3

Citation

@article{liu2023grounding,
  title={Grounding dino: Marrying dino with grounded pre-training for open-set object detection},
  author={Liu, Shilong and Zeng, Zhaoyang and Ren, Tianhe and Li, Feng and Zhang, Hao and Yang, Jie and Li, Chunyuan and Yang, Jianwei and Su, Hang and Zhu, Jun and others},
  journal={arXiv preprint arXiv:2303.05499},
  year={2023}
}