Detection Module including mmdet, mmyolo

Usage

1. Setup
2. Data Preparation
3. Execution Instructions

Setup

conda create --name openmmlab python=3.8 -y
conda activate openmmlab

conda install pytorch torchvision -c pytorch
pip install -e .

Data Preparation

Prepare COCO format dataset

coco_root_dir
    ├── annotations
    │   ├── train.json 
    │   └── val.json
    ├── train 
    │   ├── image1.jpg 
    │   ├── image2.jpg
    │        '''
    └── val
        ├── image3.jpg 
        ├── image4.jpg
             '''

you can use 'labelme_to_coco.py' to convert labelme format to coco format.

python labelme_to_coco.py root_dir classes --save_dir --split_ratio
python labelme_to_coco.py ./labelme_root_dir cat dog bird

root_dir must have labelme json files and images.

root_dir
    ├── image1.jpg
    ├── image1.json
    ├── image2.jpg
    ├── image2.json
          '''

Execution Instructions

create model

# model_type can be : mmdet, mmseg, mmpose
# mmdet-model_name can be :
# cascade_rcnn, dino-4scale, dino-5scale, rtmdet_tiny, rtmdet_x, yolox_l, yolov7_l, yolov8_l
model = create_model(model_type = "mmdet", 
                     model_name = "cascade_rcnn", 
                     data_root= '/home/inbada/mars_module/mars/mars/mmdetection/coco_format_data/', 
                     classes = ('hz', 'bz', 'chem'))

model.train()
model.test()