A PyTorch project for GigaPixel image detection.

GigaDet

GigaDet is a progressive object detection framework, which consists of a PGN module and a DecDet module.

In this repository, the features below are implemented:

• train for DecDet
• inference for DecDet
• inference for GigaDet (Depends on PGN project codes)

Usage

DecDet

Train DecDet on COCO Dataset

PYTHONPATH=gigadetect:$PYTHONPATH python example/decdet_train.py --config-file configs/GigaDet/GigaDet_COCO_Tiny.yml

Evaluate DecDet on PANDA Test Set

 PYTHONPATH=gigadetect:$PYTHONPATH python example/decdet_evaluate.py --config-file configs/GigaDet/GigaDet_COCO.yml

Some key config items in config file:

• MODEL.WEIGHTS: Specifies the pre-trained model file.
• DATASETS.VAL_ROOR: Specifies the dataset which should be evaluated.
• MODEL.CFG_FILE_PATH: Specifies the config_file (such as yolov5s.yaml) for specific arch corresponding to the loaded model.

GigaDet

GigaDetSingle Image Forward

PYTHONPATH=.:path/to/pgn/repo python example/detect_panda_image.py --image_path path/to/image --config-file path/to/config/file --save_path path/to/save/result

Evaluate the whole procedure

python example/evaluate_panda_detect_service.py --config-file path/to/config/file 

Update

• 2020-08-11: Create the repository.
• 2021-02-02: Start to refactor the repository.
• 2021-03-04: continue to refactor.