| # MILL3D Benchmark Protocol |
|
|
| MILL3D is designed for controlled benchmarking across scenes and exposure levels. |
|
|
| ## Pairing Rule |
|
|
| For a given scene, images with the same `image_name` across different `level_name` folders are paired captures of the same viewpoint. |
|
|
| ## Reference Rule |
|
|
| For each low-light sample: |
| - input image: `data/{scene_name}/{level_name}/images/{image_name}` |
| - reference image: `data/{scene_name}/GT/images/{image_name}` |
|
|
| ## Intended Evaluation Use |
|
|
| - Evaluate reconstruction quality exposure-by-exposure |
| - Compare robustness across low-light severity levels |
| - Report metrics consistently across scenes and levels |
|
|
| ## Notes |
|
|
| This dataset is intended primarily for evaluation and benchmarking, rather than as a large-scale training corpus. |
|
|