Update README.md
Browse files
README.md
CHANGED
|
@@ -12,3 +12,92 @@ pretty_name: M
|
|
| 12 |
|
| 13 |
[Project Page](https://color.cvc.uab.cat/mill/) - [Arxiv](https://arxiv.org/abs/2511.15496) - [Code](https://github.com/CVC-Color/MILL)
|
| 14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
[Project Page](https://color.cvc.uab.cat/mill/) - [Arxiv](https://arxiv.org/abs/2511.15496) - [Code](https://github.com/CVC-Color/MILL)
|
| 14 |
|
| 15 |
+
## MILL Dataset
|
| 16 |
+
The MILL dataset can be downloaded from this repository or directly from the following links.
|
| 17 |
+
|
| 18 |
+
| Split | Camera | Resolution | Download Link |
|
| 19 |
+
|----------|----------|----------|----------|
|
| 20 |
+
| MILLs-DSLR | Nikon D5200 | 600x400 | [Link](https://datasets.cvc.uab.cat/color/MILLs-DSLR.zip) (196MB) |
|
| 21 |
+
| MILLs-smartphone | Samsung Galaxy S7 | 600x400 | [Link](https://datasets.cvc.uab.cat/color/MILLs-smartphone.zip) (227MB) |
|
| 22 |
+
| MILLf-DSLR | Nikon D5200 | 2012x1340 | [Link](https://datasets.cvc.uab.cat/color/MILLf-DSLR.zip) (8.4GB) |
|
| 23 |
+
| MILLf-smartphone | Samsung Galaxy S7 | 1560x1040 | [Link](https://datasets.cvc.uab.cat/color/MILLf-smartphone.zip) (915MB) |
|
| 24 |
+
|
| 25 |
+
The MILLs splits are the low-resolution splits of the MILL dataset, while the MILLf are the high-resolution ones.
|
| 26 |
+
|
| 27 |
+
### MILL-Small
|
| 28 |
+
Both MILLs images are resized to 600x400 and follow the following data structure:
|
| 29 |
+
|
| 30 |
+
```
|
| 31 |
+
MILLs-{split}
|
| 32 |
+
|- 📁 train
|
| 33 |
+
| |- 📁 input
|
| 34 |
+
| | |- 🖼️ Scene1_a-1.png
|
| 35 |
+
| | |- 🖼️ Scene1_a-2.png
|
| 36 |
+
| | | ...
|
| 37 |
+
| | |- 🖼️ Scene1_a-10.png
|
| 38 |
+
| | |- 🖼️ Scene1_b-1.png
|
| 39 |
+
| | |- 🖼️ Scene2_a-7.png
|
| 40 |
+
| | |- 🖼️ Scene9-4.png
|
| 41 |
+
| |
|
| 42 |
+
| |- 📁 gt
|
| 43 |
+
| |- 🖼️ Scene1_a.png
|
| 44 |
+
| |- 🖼️ Scene1_b.png
|
| 45 |
+
| |- 🖼️ Scene2_a.png
|
| 46 |
+
| |- 🖼️ Scene9.png
|
| 47 |
+
|
|
| 48 |
+
|- 📁 validation
|
| 49 |
+
| ...
|
| 50 |
+
|- 📁 test
|
| 51 |
+
...
|
| 52 |
+
```
|
| 53 |
+
The first part of the filename defines the scene name while the second part of the input filenames defines the intensity level (from 1 to 10).
|
| 54 |
+
|
| 55 |
+
### MILLf-DSLR
|
| 56 |
+
The full resolution DSLR split crops the 6036x4020 original images into 9 2012x1340 images. The filenames have the grid position (row-column id) in between the scene name and intensity level.
|
| 57 |
+
|
| 58 |
+
```
|
| 59 |
+
MILLf-DSLR
|
| 60 |
+
|- 📁 train
|
| 61 |
+
| |- 📁 input
|
| 62 |
+
| | |- 🖼️ Scene1_a-00-1.png
|
| 63 |
+
| | |- 🖼️ Scene1_a-01-1.png
|
| 64 |
+
| | | ...
|
| 65 |
+
| | |- 🖼️ Scene1_a-22-1.png
|
| 66 |
+
| | |- 🖼️ Scene1_a-00-2.png
|
| 67 |
+
| | | ...
|
| 68 |
+
| | |- 🖼️ Scene1_a-00-10.png
|
| 69 |
+
| | |- 🖼️ Scene1_b-22-1.png
|
| 70 |
+
| | |- 🖼️ Scene2_a-12-7.png
|
| 71 |
+
| | |- 🖼️ Scene9-02-4.png
|
| 72 |
+
| |
|
| 73 |
+
| |- 📁 gt
|
| 74 |
+
| |- 🖼️ Scene1_a-00.png
|
| 75 |
+
| |- 🖼️ Scene1_a-01.png
|
| 76 |
+
| |- 🖼️ Scene1_a-22.png
|
| 77 |
+
| |- 🖼️ Scene1_b-22.png
|
| 78 |
+
| |- 🖼️ Scene2_a-12.png
|
| 79 |
+
| |- 🖼️ Scene9-02.png
|
| 80 |
+
|
|
| 81 |
+
|- 📁 validation
|
| 82 |
+
| ...
|
| 83 |
+
|- 📁 test
|
| 84 |
+
...
|
| 85 |
+
```
|
| 86 |
+
|
| 87 |
+
### MILLf-smartphone
|
| 88 |
+
It follows the same structure as the MILLs splits. The images are the original ones captured by the Samsung Galaxy S7 device.
|
| 89 |
+
|
| 90 |
+
## Results
|
| 91 |
+
We benchmark several state-of-the-art low-light enhancement methods on the MILL dataset and reveal significant performance variations across different intensity levels. Our proposed improvements achieve up to 10 dB PSNR improvement for DSLR and 2 dB for smartphone images on Full HD resolution. For detailed results, please refer to our paper.
|
| 92 |
+
|
| 93 |
+
## Citation
|
| 94 |
+
If you use the MILL dataset or our methods in your research, please cite our paper:
|
| 95 |
+
|
| 96 |
+
```bibtex
|
| 97 |
+
@inproceedings{pilligua2026mill,
|
| 98 |
+
title={MILL: Evaluating Low-Light Image Enhancement Across Multiple Intensity Levels},
|
| 99 |
+
author={Pilligua, Maria and Serrano-Lozano, David and Peng, Pai and Baldrich, Ramon and Brown, Michael S. and Vazquez-Corral, Javier},
|
| 100 |
+
booktitle={CVPR Findings},
|
| 101 |
+
year={2026}
|
| 102 |
+
}
|
| 103 |
+
```
|