This image showcases a detailed research poster presented at a conference. The poster, titled "TurboSL: Dense, Accurate and Neural Inverse Structured Light," focuses on a method for reconstructing 3D surfaces from a sequence of images captured under Structured Light (SL) patterns. The research has been conducted by Parsa Mirdehghan, Maxx Wu, Wenzheng Chen, and David B. Lindell and is affiliated with the University of Toronto, the Toronto Computational Imaging Group (TCIG), and the Vector Institute.

Key elements of the poster include:

1. **Objectives and Methods:**
   - The primary goal is to achieve fast and accurate 3D surface reconstruction using only three or four SL images captured at high frequencies (30-60Hz).
   - The method ensures accurate surfaces with sub-pixel accurate disparities and dense reconstructions.

2. **Pixelwise Disparity Estimation:**
   - The technique minimizes Zero-Mean Normalized Cross-Correlation (ZNCC) between projector and camera pixels.
   - Highlights the challenges in acquiring accurate surfaces with large N values and the long acquisition times needed for many patterns.

3. **Alternative Methods:**
   - Comparison between 2D MLP (Depth Map) and 3D MLP (Density Field).
   - 2D MLP fails to handle high-frequency patterns effectively, while 3D MLP offers accurate image reconstruction but poor geometry.

4. **TurboSL Neural Decoder:**
   - Describes the process of the neural decoder involving forward rendering and recasting grayscale multi-pattern SL as an inverse rendering problem.
   - Visual representations include depth maps, reflectance, normal maps, and cosine factors.

5. **TurboSL Encoder:**
   - Emphasizes the encoder's proficiency with high-frequency patterns, optimized for accuracy instead of precision.
   - Includes a depiction of the sample pattern and confusion matrix.

6. **Uncertainty Proxy by Bi-Normalization:**
   - Explains two rendering techniques: back-to-front and front-to-back rendering.

Attendees can be seen engaging with the poster, indicating active participation and interest in the research presented. The logos of the University of Toronto, TCIG, and Vector Institute are prominently displayed at the top, signifying institutional support and collaboration. The poster is mounted on a stand, and several personal items, including backpacks and a stool, are placed nearby, suggesting a busy and interactive conference environment.
Text transcribed from the image:
University of Toronto
TCIG Toronto Computational Imaging Group
Vector Institute

TurboSL: Dense, Accurate and
Neural Inverse Structure

Parsa Mirdehghan, Maxx Wu, Wenzheng Chen, David B. Lindell

Key Objective: Fast and Accurate 3D from Multi-Pattern SL

To reconstruct 3D surface from a sequence of images captured under N Structured Light (SL) patterns while pursuing:
- fast acquisition (only three or four SL images captured at 30-60Hz)
- accurate surfaces (sub-pixel-accurate disparities)
- dense reconstructions (mega-pixel-resolution depths)

Pixelwise Disparity Estimation

- Aims to minimize the Zero-Mean Normalized Cross-Correlation (ZNCC) between projector & camera pixels.
- Yields accurate surfaces given large N
- requires long acquisition time to capture many patterns
- Quality degrades significantly with small N

Alternatives for 3D from SL: Scene Representation with MLP

2D MLP - Depth Map
fails to handle high-frequency patterns

3D MLP - Density Field
accurate image reconstruction, but poor geometry

Our Approach: TurboSL by Neural Inverse Structured Light

By recasting grayscale multi-pattern SL as an inverse rendering problem, 3 or four projection patterns are enough for detailed geometric and reflectance information.

depth map
reflectance 
SL image 3
normal map
cosine factor

1. TurboSL Neural Decoder
forward rendering
reflectance cosine factor reflectance cosine factor
camera projector

2. TurboSL Encoder
TurboSL works best with high-frequency patterns optimized for accuracy instead of precision [1]
sample pattern confusion matrix

3. Uncertainty Proxy by Binding

back-to-front rendering
Ncam proj
Ncam proj
front-to-back rendering