Dataset & LoRA Training Pipeline for Room Emptying / Object Removal (Image-to-Image)

A community member is seeking guidance on dataset sources and training architecture for a project that empties indoor spaces and removes specific objects from images using provided object masks. The poster asks whether paired furnished-room/empty-room datasets such as InteriorNet, Structured3D, or Matterport3D suit an image-to-image or diffusion pipeline, and whether fine-tuning a LoRA on an inpainting model like SDXL Inpainting or Flux Inpainting with mask-guided conditioning is preferable to training a custom ControlNet. The request also asks for relevant papers, datasets, and codebases.

Hi Community, I am working on a project aimed at emptying indoor spaces removing all furniture/decor as well as removing specific individual objects from an image using provided object masks. I plan to fine-tune a model or train a LoRA / ControlNet adapter , but I need guidance on dataset sources and the ideal training architecture. Here are my specific questions: Dataset Sources for Empty Room Pairs: Where can I find datasets containing paired images of Furnished Room ↔ Empty Room suitable for training an image-to-image or diffusion pipeline? Are there existing synthetic or 3D-rendered datasets e.g., InteriorNet, Structured3D, Matterport3D recommended for this specific paired generation task? Single Object Removal & Inpainting: For the use case of removing a single specific object where I already have the object mask , what is the recommended architecture? Is fine-tuning a LoRA on an Inpainting model like SDXL Inpainting or Flux Inpainting using mask-guided conditioning preferable over training a custom ControlNet? Any pointers to relevant papers, datasets, or codebases would be greatly appreciated Thanks