venkatsubra/Uniformat

The official implementation is available on GitHub.

Zero-Shot Depth from Defocus

Yiming Zuo* · Hongyu Wen* · Venkat Subramanian* · Patrick Chen · Karhan Kayan · Mario Bijelic · Felix Heide · Jia Deng

(*Equal Contribution)

Princeton Vision & Learning Lab (PVL)

Paper · Project

---

Overview

We use the Gaussian PSF and the photographer in-the-loop FD to synthesize the focus stack from the RGBD input on iBims[20], DIODE[42] and HAMMER[18].

These datasets often have incomplete depth maps, either due to occlusions, low-reflectance surfaces not capturable by depth sensors, or regions too far away. Those missing regions cause trouble when synthesizing the defocus image, as the synthesized image is unsmooth and contains a lot of artifacts if we naively assume a constant depth value in the missing regions. Therefore, we use an off-the-shelf depth completion network OMNI-DC[56] to fill the missing regions. Note that the filled depth values are only used to synthesize the focus stack but does not affect the evaluation metrics, because we only compute the metrics on the valid depth pixels in ground-truth.

— Paper (arXiv)

---

Unzipped File Format

After unzipping DefocusUniformat.zip, the expected directory structure is:

tmp_defocus_uniformat_upload/
├── diode/
│   ├── diode_indoor_v2/
│   │   ├── 000000.npy
│   │   │   Keys:
│   │   │   - gt: raw depth
│   │   │   - depth_filled: OMNI-DC filled depth
│   │   │   - rgb: RGB image
│   │   │   - valid_mask: valid mask for the depth map; pixels filled by OMNI-DC are marked as false.
│   │   │   - K: camera intrinsics matrix
│   │   ├── 000001.npy
│   │   └── ...
│   └── diode_outdoor_v2/
│       ├── 000000.npy
│       ├── 000001.npy
│       └── ...
└── ibims/
    ├── 000000.npy
    ├── 000001.npy
    └── ...