Uncertainty_diffusion

Coding environment installation

create a clean environment udiffusion

git clone https://github.com/giannifranchi/Uncertainty_diffusion
cd Uncertainty_diffusion
git checkout develop # if using the develop otherwise ignore
conda create -n udiffusion python=3.10
conda activate udiffusion
pip install -r requirements.txt

For molmo evaluation create a clean environment udiffusion

git clone https://github.com/giannifranchi/Uncertainty_diffusion
cd Uncertainty_diffusion
git checkout develop # if using the develop otherwise ignore
conda create -n udiffusionmolmo python=3.10
conda activate udiffusionmolmo
pip install -r requirements_molmo.txt

Extract Prompt Dataset

bash scripts/extract_prompt_dataset.sh

Download models for toy exemple

bash scripts/download_toy_models.sh
bash scripts/download_toy_vae_from_pixart.sh

bash scripts/run.sh

Exemple on how to launch the evalutation

python compute_results_per_model.py --input_dir prompt_dataset --label2 Microscopic --label1 Normal  --model pixart --save_path pipe_get_confident_score

run evaluation with molmo

An example script using sdxs with molmo is in scripts/sdxs_molmo.sh

bash scripts/sdxs_molmo.sh