Research map / Architectures and efficiency / CNN design
ASP Vision: Optically Computing the First Layer of Convolutional Neural Networks Using Angle Sensitive Pixels
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace the CNN's first layer with ASP optical edge filtering | LeNet on MNIST · MNIST test | accuracy: 99.12% -> 99.04% | No clear effect |
| Replace the CNN's first layer with ASP optical edge filtering | NiN · CIFAR-10 grayscale | accuracy: 84.90% -> 81.8% | Hurt |
| Replace the CNN's first layer with ASP optical edge filtering | NiN · CIFAR-100 grayscale | accuracy: 55.60% -> 50.9% | Hurt |
| Replace the CNN's first layer with ASP optical edge filtering | VGG-M-128 on PF-83 · PF-83 | accuracy: 65.67% -> 66.8% | No clear effect |
| Use ASP optical edge filtering instead of the conventional sensor and learned first layer | LeNet prototype · MNIST hardware, regular dataset | validation accuracy: 91.26% with σ = 2.77% -> 86.7% with σ = 3.75% | Hurt |
| Augment the hardware training data with linear shifts and rotations | ASP Vision prototype · MNIST hardware | validation accuracy: 86.7% with σ = 3.75% -> 94.61% with σ = 0.68% | Helped |
| Use ASP optical edge filtering instead of the conventional sensor and learned first layer | NiN prototype · Face identification hardware, regular dataset | validation accuracy: 93.53% with σ = 8.37% -> 82.87% with σ = 18.12% | Hurt |
| Augment the hardware training data with linear shifts and rotations | ASP Vision prototype · Face identification hardware | validation accuracy: 82.87% with σ = 18.12% -> 94.18% with σ = 5.04% | Helped |
| Use ASP optical edge filtering | ASP Vision versus LeNet · MNIST at 9dB SNR | accuracy: 42.6% -> 38.6% | Hurt |
| Use ASP optical edge filtering | ASP Vision versus LeNet · MNIST at 12dB SNR | accuracy: 83.6% -> 77.9% | Hurt |
| Vary ASP angular frequency and grating orientation (ablation) | ASP Vision on MNIST · MNIST | accuracy | No clear effect |
| Use 6 randomized ASP filters across 100 simulations (ablation) | ASP Vision on MNIST · MNIST | error: 1.13% error with a standard deviation of 0.13% | No clear effect |
Closest papers by meaning
- Efficient Neural Vision Systems Based on Convolutional Image Acquisition (CVPR 2020 · similarity 0.86)
- Neural Tangent Knowledge Distillation for Optical Convolutional Networks (NeurIPS 2025 · similarity 0.84)
- VISTREAM: Improving Computation Efficiency of Visual Streaming Perception via Law-of-Charge-Conservation Inspired Spiking Neural Network (CVPR 2025 · similarity 0.83)
- AdaptiveISP: Learning an Adaptive Image Signal Processor for Object Detection (NeurIPS 2024 · similarity 0.83)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.82)
- QuartDepth: Post-Training Quantization for Real-Time Depth Estimation on the Edge (CVPR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.