ReconfigISP [arXiv][Project Page]
This repo contains the main code of ReconfigISP.
The OnePlus dataset is available here.
training without proxy tuning
python -m torch.distributed.launch --nproc_per_node=4 --master_port=$RANDOM train.py --opt options/train/{YOUR_OPTION_FILE}.yml --launcher pytorch
training with proxy tuning
python -m torch.distributed.launch --nproc_per_node=4 --master_port=$RANDOM train_ft.py --opt options/train/{YOUR_OPTION_FILE}.yml --launcher pytorch
normal inference
python test.py --opt options/test/{YOUR_OPTION_FILE}.yml
split into patches for inference (save GPU memory)
python test_split.py --opt options/test/{YOUR_OPTION_FILE}.yml
split into patches to test object detection
python test_yolo_split.py --opt options/test/{YOUR_OPTION_FILE}.yml
The implementation of architecture search is borrowed from DARTS.
The images are annotated by labelme.