Related Experiment Video
Updated: Jun 21, 2026
09:23
Harmonic Nanoparticles for Regenerative Research
Published on: May 1, 2014
IDNet: An inception-like deformable non-local network for projection compensation over non-flat textured surfaces
Yuqiang Zhang1, Huamin Yang1, Cheng Han1
1School of Computer Science and Technology, Changchun University of Science and Technology, Changchun, Jilin, China.
Plos One
|May 20, 2025
Summary
IDNet enhances projector compensation on complex surfaces using multi-scale deformable convolutions and attention mechanisms. This innovative framework improves geometric and photometric accuracy, especially in challenging image edge regions.
Area of Science:
- Computer Vision
- Computational Imaging
- Image Processing
Background:
- Projector compensation on non-flat, textured surfaces is difficult.
- Conventional methods struggle with geometric distortions, especially at image edges.
- Complex surface geometry limits accuracy in standard image compensation techniques.
Purpose of the Study:
- To introduce IDNet, a novel framework for robust projector compensation.
- To address limitations of existing methods in handling geometric transformations.
- To improve both geometric and photometric compensation accuracy on challenging surfaces.
Main Methods:
- Utilizing multi-scale receptive feature extraction modules.
- Employing multi-scale deformable convolutions with adaptive sampling and offsets.
- Integrating non-local attention mechanisms for global contextual information capture.
Main Results:
- IDNet demonstrates effective compensation on non-flat, textured surfaces.
- The framework shows enhanced processing in intricate distortion regions.
- Achieved comparable performance to existing methods, excelling in complex edge regions.
Conclusions:
- IDNet provides a significant advancement in projector compensation technology.
- The proposed method offers improved accuracy for geometrically complex projections.
- IDNet is particularly effective for challenging non-flat and textured surfaces.
