Retinexformer · ICCV 2023

Last 30 days—Today—

Upload a dark photo, choose a checkpoint, and reveal the details. Compare the input and enhanced image side by side, then download the result.

Low-light image

Enhanced result

Example images — click a thumbnail, then Enhance image

Model weights

Retinexformer .pth files in pretrained_weights.

Output brightness

1 = model output; >1 brighter, <1 darker. After enhancement, release to update instantly.

0.1 3

Gamma curve

1 = original input; lower values brighten the input.

0.1 5

Maximum image size

Longest edge. Larger photos are resized before inference.

Self-ensemble (8 passes)

Average flipped / rotated predictions; takes longer.

Inference details

Selected weights, CPU/GPU, elapsed time, input/output dimensions and enhancement settings.

Device: cuda:0 · Tip: Start with Gamma = 1.0. Use the download icon on the result to save a PNG.