DiffusionLight uses a diffusion model to inpaint a chrome ball into the input image and unwarp it to an environment map. It also fine-tunes an LDR diffusion model to perform exposure bracketing for HDR light estimation.
DiffusionLight uses a diffusion model to inpaint a chrome ball into the input image.
DiffusionLight unwraps the inpainted chrome ball to an environment map.
DiffusionLight fine-tunes an LDR diffusion model to perform exposure bracketing for HDR light estimation.
DiffusionLight can generate multiple plausible chrome balls by varying the initial noise map of diffusion sampling.
DiffusionLight can seamlessly insert 3D objects into an existing photograph using its environmental light estimates.
Estimate lighting for indoor and outdoor scenes.
Estimate lighting for close-up shots and paintings.
Estimate lighting for photos of human faces.
Insert 3D objects into an existing photograph.
Generate multiple plausible chrome balls for lighting estimation.
Input an image into the DiffusionLight model.
The model will inpaint a chrome ball into the input image.
The model will unwarp the inpainted chrome ball to an environment map.
The model will fine-tune an LDR diffusion model to perform exposure bracketing for HDR light estimation.
The model will generate multiple plausible chrome balls for lighting estimation.