InfiniDepth: Arbitrary-Resolution and Fine-Grained Depth Estimation with Neural Implicit Fields

InfiniDepth introduces a novel depth representation that models depth as neural implicit fields, enabling arbitrary-resolution and fine-grained monocular depth estimation.

The method uses a Vision Transformer encoder to extract multi-scale features, which are queried at continuous 2D coordinates via bilinear interpolation and fused hierarchically through a lightweight MLP decoder.

This allows direct prediction at any resolution, overcoming the limitations of discrete grid-based methods.

The paper also proposes an Infinite Depth Query strategy that allocates sub-pixel query budgets based on surface area, producing uniformly distributed 3D points that improve novel view synthesis under large viewpoint shifts.

To evaluate high-resolution and fine-detail performance, the authors curate Synth4K, a 4K synthetic benchmark from five games, with high-frequency masks for targeted detail assessment.

Experiments show state-of-the-art results on Synth4K and real-world benchmarks for both relative and metric depth estimation, with significant gains in fine-detail regions.

The method also enhances single-view novel view synthesis, reducing holes and artifacts compared to prior approaches.