Deshadow-Anything: When Segment Anything Model Meets Zero-shot shadow removal

  • 2024-01-03 02:01:09
  • Xiao Feng Zhang, Tian Yi Song, Jia Wei Yao
  • 0

Abstract

Segment Anything (SAM), an advanced universal image segmentation modeltrained on an expansive visual dataset, has set a new benchmark in imagesegmentation and computer vision. However, it faced challenges when it came todistinguishing between shadows and their backgrounds. To address this, wedeveloped Deshadow-Anything, considering the generalization of large-scaledatasets, and we performed Fine-tuning on large-scale datasets to achieve imageshadow removal. The diffusion model can diffuse along the edges and textures ofan image, helping to remove shadows while preserving the details of the image.Furthermore, we design Multi-Self-Attention Guidance (MSAG) and adaptive inputperturbation (DDPM-AIP) to accelerate the iterative training speed ofdiffusion. Experiments on shadow removal tasks demonstrate that these methodscan effectively improve image restoration performance.

 

Quick Read (beta)

loading the full paper ...