GroundingSeg

This repo is not refactored yet， not ready for out-of-box usage currently!!!

Installation

This project is modified from mmdetection. Follow the steps:

1. Create a conda env (Strongly suggested!) and activate, install Python and PyTorch. Our project is developed under Ubuntu 22.04 + Python=3.10 + torch=2.0.1 + cu118
2. Install mmdetection. Our project is modified from mmdet==3.2.0

pip install -U openmim==0.3.9
mim install mmengine==0.10.4
mim install mmcv==2.1.0

3. Install GroundingSeg

git clone https://github.com/ABC67876/GroundingSeg.git
cd mmdetection
pip install -v -e .
# "-v" means verbose, or more output
# "-e" means installing a project in editable mode,
# thus any local modifications made to the code will take effect without reinstallation.

pip install -r requirements.txt
# Not yet filtered for the minimum version

4. visit the https://github.com/ABC67876/GroundingSeg/tree/main/patch to fix some issues.

Usage

Generally, run the https://github.com/ABC67876/GroundingSeg/blob/main/mmdetection/tools/train.py and https://github.com/ABC67876/GroundingSeg/blob/main/mmdetection/tools/test.py

For each dataset, we have a config file for detection and segmentation respectively.

Data format should be COCO for detection and COCO panoptic for segmentation.

We are currently working on this repo for better documents. This part is under progress.

We will provide scripts for data generating, pretrain weights, and refactor this project for easier usage.