Research map / Architectures and efficiency / CNN design
A Camera That CNNs: Towards Embedded Neural Networks on Pixel Processor Arrays
Laurie Bose, Jianing Chen, Stephen J. Carey, Piotr Dudek, Walterio Mayol-Cuevas
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| use non-zero ternary threshold (ternary weights) instead of threshold 0 (binary weights) (ablation) | Ternary CNN on SCAMP5 · MNIST (SCAMP5) | accuracy and convolution time: best accuracy 94.2% at threshold 0.2 | Improved |
| raise ternary threshold from 0 to 0.99 (sparser weights) (ablation) | Ternary CNN on SCAMP5 · MNIST (SCAMP5) | frame rate (fps): 135 -> 210 fps | Improved |
| raise ternary threshold to 0.99 (only a couple of non-zero weights per kernel) (ablation) | Ternary CNN on SCAMP5 · MNIST (SCAMP5) | accuracy (%): 94.2% -> 70% | Hurt |
| deploy on SCAMP5 sensor instead of PC evaluation | MNIST CNN · MNIST | accuracy (%): 95.4% -> 94.2% | Hurt |
| use non-zero ternary threshold instead of binary weights (ablation) | Ternary CNN on SCAMP5 · Car tracking | localization error | Improved |
| raise ternary threshold from 0 to 0.99 (ablation) | Ternary CNN on SCAMP5 · Car tracking | frame rate (fps): 140 -> 250 fps | Improved |
| raise ternary threshold above 0.5 (ablation) | Ternary CNN on SCAMP5 · Car tracking | localization error: roughly constant for thresholds 0.2 to 0.5, then exponential increase | Hurt |
Closest papers by meaning
- Fully Embedding Fast Convolutional Networks on Pixel Processor Arrays (ECCV 2020 · similarity 0.94)
- Focal Plane Visual Feature Generation and Matching on a Pixel Processor Array (ICCV 2025 · similarity 0.86)
- Descriptor-In-Pixel : Point-Feature Tracking For Pixel Processor Arrays (CVPR 2025 · similarity 0.85)
- PixelRNN: In-pixel Recurrent Neural Networks for End-to-end-optimized Perception with Neural Sensors (CVPR 2024 · similarity 0.83)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.83)
- An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels (ICLR 2025 · similarity 0.83)
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.
