A Review of Diffusion Models for Image Inpainting
Abstract
Inpainting is an image processing task which involves filling the missing parts of an image. The research in image inpainting has evolved from traditional Patch based methods to Deep Learning techniques. Initially, CNNs were used for filling the missing parts of an image. Later more advanced methods such as Generative Adversarial Networks (GANs) and Diffusion based models were developed. Unlike GAN based methods, diffusion models use iterative denoising process. This process allows generation of complex textures. This study presents a comparative study of diffusion based methods. The comparison is done based on parameters such as architecture, conditioning methods and guidance strategies. We discuss some popular diffusion based models such as RePaint, Stable Diffusion Inpainting, LatentPaint, RAD and Global Structure Guided Diffusion. The study highlights the role of mask-aware conditioning, latent diffusion and attention based guidance in generating high quality output. The study also highlights the trade-offs between efficiency, flexibility and output quality. Most of the models are able to optimise one or two factors but not all three. Furthermore, we explore how diffusion can be combined with GANs, transformers to achieve multi-modal conditioning. This improves the flexibility of the model to produce uses guided output. Finally, the challenges and future direction is discussed highlighting the need for efficient models, video inpainting and 3D image inpainting.
Read the paper
Related studies
- Learning from First-Generation Qualitative Approaches in the IS Discipline: An Evolutionary View and Some Implications for Authors and Evaluators (PART 1/2)
- Commentary on the clinical and preclinical dosage limits of interstitially administered magnetic fluids for therapeutic hyperthermia based on current practice and efficacy models
- Prosthesis: New Technological Opportunities and Innovative Biomedical Devices
- The history and future of neural modeling for cochlear implants
- Ryanodine receptors: waking up from refractoriness
- Novel Transgenic Rabbit Model Sheds Light on the Puzzling Role of Matrix Metalloproteinase-12 in Atherosclerosis
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.