16bit-from-8bit Image Reconstruction Model
This model reconstructs 16‑bit per channel images from standard 8‑bit input images. It is trained on paired 8‑bit and 16‑bit data and optimized to preserve color fidelity and high‑frequency detail.
Median MAE: 470
Weighted Mean MAE: 747
Dataset
- Total images: 54,580
- RAW patch images: 46,000 (~10 GB)
- 48‑bit synthetic images: 8,580 (~2 GB)
Evaluation Summary
| MAE Range | Count | Percent (%) | Accuracy Comment |
|---|---|---|---|
| 0–200 | 1,698 | 3.11% | Excellent: Almost perfect reconstruction. |
| 201–400 | 20,344 | 37.27% | Very good: Minor differences; mostly imperceptible. |
| 401–500 | 7,556 | 13.84% | Good: Slight color shifts in uniform areas. |
| 501–1000 | 15,717 | 28.79% | Moderate: Noticeable deviations. |
| 1001–2000 | 5,666 | 10.38% | Low: Significant perceptual errors. |
| >2000 | 3,599 | 6.59% | Poor: Highly visible errors. |
Intended Use
Primary Use Cases
- Reconstruction of 16‑bit per channel images from 8‑bit input.
- HDR post‑processing and enhancement.
- Archival and art restoration workflows.
Not Intended For
- Lossless scientific measurement or precision tasks.