sEDICT (stochastic EDICT)

The baseline model is EDICT: Exact Diffusion Inversion via Coupled Transformation in this repository.

Experimental Results

Run Example

# reference-based image generation
python -u s_EDICT.py --b_prompt 'A dog' --t_prompt 'A dog' --s_i 0.7 --s_d 0.8 --seed 1 --scale 0.2

# image editing
python -u s_EDICT.py --b_prompt 'A dog' --t_prompt 'A golden retriever' --s_i 0.7 --s_d 0.8 --seed 1 --scale 0.2

Evaluation

python -u qualitative_evaluation.py --reference_img_path REF_PATH --generated_img_path GEN_PATH