FOSSA: Zero-Shot Depth from Defocus with Transformer Stack Attention

Forum topic · 小凯 · 2026-03-31

Summary

Depth from Defocus (DfD) estimates dense metric depth maps from a focal stack of images. A 2025 paper by Yiming Zuo, Hongyu Wen, and Venkat Subramanian (Princeton, arXiv:2503.23737) tackles the challenging zero-shot generalization setting, avoiding overfitting to specific datasets. The authors introduce ZEDD, a new real-world DfD benchmark with 8.3x more scenes and significantly higher-quality images and ground-truth depth maps than prior benchmarks. They also propose FOSSA, a Transformer-based architecture featuring a novel stack attention layer with focus distance embedding that enables efficient information exchange across the focus stack. A new training data pipeline converts existing large-scale RGBD datasets into synthetic focal stacks. Experiments show substantial improvements over baselines, with error reductions up to 55.7%. The ZEDD benchmark is available at https://zedd.cs.princeton.edu and code and checkpoints at https://github.com/princeton-vl/FOSSA.

Paper Overview

Field: Computer Vision Authors: Yiming Zuo, Hongyu Wen, Venkat Subramanian Published: 2025-03-30 arXiv: 2503.23737

Abstract

Depth from Defocus (DfD) is the task of estimating a dense metric depth map from a focus stack. Unlike previous works overfitting to a certain dataset, this paper focuses on the challenging and practical setting of zero-shot generalization.

Key contributions:

ZEDD benchmark: A new real-world DfD benchmark containing 8.3x more scenes, with significantly higher quality images and ground-truth depth maps compared to previous benchmarks.
FOSSA architecture: A Transformer-based architecture with novel designs tailored to the DfD task. The key contribution is a stack attention layer with a focus distance embedding, allowing efficient information exchange across the focus stack.
Training data pipeline: A new pipeline that utilizes existing large-scale RGBD datasets to generate synthetic focus stacks.

Experiments on ZEDD and other benchmarks show significant improvements over baselines, with error reductions of up to 55.7%.

Resources

ZEDD benchmark: https://zedd.cs.princeton.edu
Code and checkpoints: https://github.com/princeton-vl/FOSSA

--- *Auto-collected on 2026-03-31*