ziplab/Awesome-Feed-Forward-3D

What it solves

Researchers and practitioners working on 3D scene modeling from images often face a fragmented landscape: many papers, scattered codebases, and unclear connections between methods, datasets, and real-world uses. This project addresses that by gathering the key work in feed-forward 3D modeling into one structured, up-to-date reference list.

How it works

The project is an organized list (an "awesome" list) that categorizes the field into three main parts: research directions, datasets and benchmarks, and applications. Each category is further broken down into sub-topics — for example, feature enhancement, cross-view fusion, geometry-aware improvement, temporal models, and more. Each entry links to its paper, code, or project page, making it easy to jump from a high-level overview to the actual resources.

Who it's for

It is aimed at researchers, students, and developers who want to explore or stay current with feed-forward 3D reconstruction methods, whether they are looking for a specific technique, a benchmark dataset, or an application area like autonomous driving or robotics.

Highlights

• Covers a broad set of research directions, including advanced encoding architectures, cross-view fusion, geometry-aware improvements, and temporal modeling.
• Includes key datasets and benchmarks grouped by geometry-oriented, visual-oriented, and mixed categories.
• Lists applications across autonomous driving, robotics, SLAM, scene understanding, and video generation.
• Each entry provides direct links to papers, code, and project pages, saving hours of searching.