Evalution_Scores_Recording.md

Hyperparameter ablation study of quantitative comparison.

4. 3.0 NPI vs. stage123

{'lpips': tensor(0.0356), 'ssim': tensor(0.9806), 'clip_score': tensor(21.3158, grad_fn=), 'psnr': tensor(40.5955), 'mse': 0.00013039794848209831} {'lpips': tensor(0.6393), 'ssim': tensor(0.5821), 'clip_score': tensor(21.2223, grad_fn=), 'psnr': tensor(11.8366), 'mse': 0.07992389089549365}

5.0 NPI vs. 123

{'lpips': tensor(0.0355), 'ssim': tensor(0.9806), 'clip_score': tensor(21.3157, grad_fn=), 'psnr': tensor(40.6037)} {'lpips': tensor(0.6393), 'ssim': tensor(0.5821), 'clip_score': tensor(21.2223, grad_fn=), 'psnr': tensor(11.8366)}

7.0 NPI vs. 123

{'lpips': tensor(0.0356), 'ssim': tensor(0.9806), 'clip_score': tensor(21.3158, grad_fn=), 'psnr': tensor(40.5959)} {'lpips': tensor(0.6393), 'ssim': tensor(0.5821), 'clip_score': tensor(21.2223, grad_fn=), 'psnr': tensor(11.8366)}

1.0 NTI vs. 123

{'lpips': tensor(0.0356), 'ssim': tensor(0.9806), 'clip_score': tensor(21.3155, grad_fn=), 'psnr': tensor(40.5930), 'mse': 0.0001304753707607734} {'lpips': tensor(0.6393), 'ssim': tensor(0.5821), 'clip_score': tensor(21.2223, grad_fn=), 'psnr': tensor(11.8366), 'mse': 0.07992389089549365}

3.0 NTI vs. 123

{'lpips': tensor(0.0353), 'ssim': tensor(0.9806), 'clip_score': tensor(21.3161, grad_fn=), 'psnr': tensor(40.6459), 'mse': 0.0001290986220355726} {'lpips': tensor(0.6393), 'ssim': tensor(0.5821), 'clip_score': tensor(21.2223, grad_fn=), 'psnr': tensor(11.8366), 'mse': 0.07992389089549365}

5.0 NTI vs. 123

{'lpips': tensor(0.0353), 'ssim': tensor(0.9806), 'clip_score': tensor(21.3161, grad_fn=), 'psnr': tensor(40.6484), 'mse': 0.00012907061201089613} {'lpips': tensor(0.6393), 'ssim': tensor(0.5821), 'clip_score': tensor(21.2223, grad_fn=), 'psnr': tensor(11.8366), 'mse': 0.07992389089549365}

7.0 NTI vs. 123

{'lpips': tensor(0.0353), 'ssim': tensor(0.9806), 'clip_score': tensor(21.3161, grad_fn=), 'psnr': tensor(40.6427), 'mse': 0.00012912355525285406} {'lpips': tensor(0.6393), 'ssim': tensor(0.5821), 'clip_score': tensor(21.2223, grad_fn=), 'psnr': tensor(11.8366), 'mse': 0.07992389089549365}

ours123_new vs. SDXL

SDXL： {'lpips': tensor(0.1269), 'ssim': tensor(0.9048), 'clip_score': tensor(21.3172, grad_fn=), 'psnr': tensor(33.4993), 'mse': 0.009015853739043267}

CUDA_VISIBLE_DEVICES=7 python end2end_inv.py
--input_image "images/gnochi_mirror.jpeg"
--prompt_str "a cat sitting on a wooden counter looking at itself in the mirror"
--results_folder "output/all_imgs_inversion_in_prompt_file"
--guidance_1 5.0
--is_NPI