Fine Tuning TransLab

[true-real-michael]

This branch is under construction and only has basic fine-tuning logic

Currently fine tuning requires running without a docker image

1. Create directories ./datasets/Trans10K/images and ./datasets/Trans10K/masks and place images and masks to the corresponding directories
2. Install requirements:

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
pip install gdown thop
gdown 1shxxcNHmpu8TmLosT1qmFnojqwQzM22e
cp -r ../common .

3. Run the fine-tuner:

python fine_tuner.py --config-file configs/trans10K/translab_bs4.yaml TRAIN.MODEL_SAVE_DIR workdirs/translab_bs4 TRAIN.PRETRAINED_MODEL_PATH 16.pth

For some reason, it does not work on my my machine with a discrete graphics card, and it works on a machine with a mac m1 processor.