bcmi/Awesome-Object-Insertion

What it solves

Creating a realistic composite image by inserting a foreground object into a background is hard because the two sources often clash in appearance, geometry, and meaning. This project collects the resources needed to tackle that problem — papers, datasets, and a toolbox — so researchers and developers can find everything in one place.

How it works

It is an "Awesome" list, meaning a hand-curated directory of links. The README organizes content into sections: an online demo, a survey paper, a toolbox (libcom) that bundles 10+ image composition functions, and topical sub-lists for tasks like blending, harmonization, shadow generation, object placement, spatial transformation, occlusion, foreground search, and generative composition. Each sub-list links to a dedicated curated repository. It also points to datasets for each task and an evaluation tool for composite images.

Who it's for

Researchers, students, and practitioners working on image composition, object insertion, or related computer vision tasks. It is also useful for anyone who wants a quick starting point for tools and benchmarks in this area.

Highlights

• Links to an online demo built on the libcom toolbox for hands-on experimentation
• Includes a comprehensive survey paper on deep image composition
• Provides a toolbox (libcom) that aims to solve image composition problems with a simple import libcom
• Curates 9 sub-topics, each with its own resource list (blending, harmonization, shadow generation, reflection generation, placement, spatial transformation, occlusion, foreground search, generative composition)
• Aggregates datasets and evaluation tools across all sub-tasks