# EM3D

**Repository Path**: sheldonhe/EM3D

## Basic Information

- **Project Name**: EM3D
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-06-20
- **Last Updated**: 2025-06-20

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

The MATLAB codes allow detailed and precise 3D reconstruction of an experimental single-shot image of a nearly transparent glass beaker.

3D vision is of paramount importance for machine intelligence and many other application scenarios. Despite much recent progress, most 3D imaging hardwares remain bulky and complicated, and provide image pixels much fewer than their 2D counterparts. Moreover, there are many well-known corner cases that existing 3D imaging solutions may fail. Here, we propose and experimentally realize an extended monocular 3D imaging (EM3D) framework that can reconstruct detailed 3D surfaces with accurate absolute depth for various scenes that are traditionally challenging, including those with low-texture, being highly reflective, or nearly transparent. The monocular camera obtains the 3D information in a snapshot via the multi-stage fusion of two depth cues: 1) absolute yet incomplete depth cue from a depth-dependent point-spread function generated by a diffractive optical element; 2) ambiguous yet detailed depth cue from a polarization image sensor. We further combine depth and polarization information obtained from the monocular camera for material discrimination and demonstrate robotic target recognition and face anti-spoofing. The image reconstruction algorithm is physically-interpretable and does not rely on any prior information or training dataset. Such a straightforward yet powerful architecture may facilitate the deployment of 3D imaging in much more diverse scenarios than ever before.
![image](https://github.com/user-attachments/assets/f73e08f9-99cd-4144-b1f2-fa8e89d2f298)

EM3D_experiment.m - Detailed and precise 3D reconstruction from the experimental image