Related Experiment Video
Updated: Oct 12, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.0K
Incoherent reconstruction-free object recognition with mask-based lensless optics and the Transformer
Optics Express
|November 23, 2021
Summary
This study introduces a novel method for object recognition using mask-based lensless cameras. By performing recognition directly on encoded patterns, it bypasses image reconstruction, saving resources and improving accuracy.
Area of Science:
- Optics
- Computer Vision
- Machine Learning
Background:
- Lensless cameras offer advantages in size and cost over traditional cameras.
- Image reconstruction is computationally intensive and can introduce artifacts.
- Object recognition is a key application for compact imaging systems.
Purpose of the Study:
- To develop a method for direct object recognition on encoded patterns from mask-based lensless cameras.
- To investigate the potential of Transformer-based architectures for this task.
- To evaluate the performance of the proposed system on benchmark datasets.
Main Methods:
- A mask-based lensless camera system was employed to capture encoded scene patterns.
- A simplified Transformer-based neural network was designed for direct pattern recognition.
- The system was tested on the Fashion MNIST and cats-vs-dogs datasets.
Main Results:
- The system achieved 91.47% accuracy on the Fashion MNIST dataset.
- The system obtained a 96.64% ROC AUC on the cats-vs-dogs dataset.
- The feasibility of physical object recognition using this approach was demonstrated.
Conclusions:
- Direct object recognition on encoded patterns from lensless cameras is feasible and efficient.
- Transformer-based architectures are effective for extracting features from encoded patterns.
- This approach offers a computationally efficient alternative for inference tasks with lensless cameras.
