ByteMorph-Eval

1. Save your generation results as the GT format (Left: Input; Right: Output)

1.1 Specify the generation root, gt root and metrics output directory in eval_clip_score.sh and eval_gpt_score.sh.

2. Install related packages

cd ./ByteMorph-Eval
conda create -n bytemorph_eval python=3.10
conda activate bytemorph_eval
pip3 install -r requirements.txt

3. Run the evaluation script

bash eval_clip_score.sh

The metric results will be save in {OUTPUT_SCORE_DIR}/clip_score/{EDITING_TYPE}_avg.json.

4. Additional steps for evaluation with VLM (GPT)

4.1 Configure your API key and model version in gpt_eval.py

4.2 After configuration, run the following script:

bash eval_gpt_score.sh

The final score will be save in {OUTPUT_SCORE_DIR}/gpt_score/{EDITING_TYPE}/average_score.json.