InfiniDepth: Arbitrary-Resolution and Fine-Grained Depth Estimation with Neural Implicit Fields
Hao Yu, Haotong Lin, Jiawei Wang, Jiaxin Li, Yida Wang, Xueyang Zhang, Yue Wang, Xiaowei Zhou, Ruizhen Hu, Sida Peng
2026-01-07
Summary
This paper introduces a new way to estimate depth in images, moving beyond traditional methods that only predict depth at specific points in the image.
What's the problem?
Current depth estimation techniques work by calculating depth values on a grid, like a checkerboard over the image. This limits how detailed the depth map can be and makes it difficult to create depth maps with very high resolution or capture fine details because you're stuck with the resolution of the grid. Essentially, they can't easily zoom in and get precise depth information everywhere.
What's the solution?
The researchers developed a method called InfiniDepth that represents depth not as a grid, but as a continuous field. Imagine it like a smooth surface instead of a bumpy one. They use a neural network to learn this surface, allowing them to ask for the depth at *any* point in the image, not just the grid points. This allows for much higher resolution and more detailed depth maps. They also created a new, high-quality dataset from video games to test their method.
Why it matters?
This is important because more accurate and detailed depth maps have many applications, like creating realistic 3D models from images, improving virtual reality experiences, and helping robots understand the world around them. InfiniDepth performs better than existing methods, especially when it comes to capturing small details, and also improves the quality of creating new views of a scene.
Abstract
Existing depth estimation methods are fundamentally limited to predicting depth on discrete image grids. Such representations restrict their scalability to arbitrary output resolutions and hinder the geometric detail recovery. This paper introduces InfiniDepth, which represents depth as neural implicit fields. Through a simple yet effective local implicit decoder, we can query depth at continuous 2D coordinates, enabling arbitrary-resolution and fine-grained depth estimation. To better assess our method's capabilities, we curate a high-quality 4K synthetic benchmark from five different games, spanning diverse scenes with rich geometric and appearance details. Extensive experiments demonstrate that InfiniDepth achieves state-of-the-art performance on both synthetic and real-world benchmarks across relative and metric depth estimation tasks, particularly excelling in fine-detail regions. It also benefits the task of novel view synthesis under large viewpoint shifts, producing high-quality results with fewer holes and artifacts.