Research map / Architectures and efficiency / Neural architecture search
SACoD: Sensor Algorithm Co-Design Towards Efficient CNN-Powered Intelligent PhlatCam
Yonggan Fu, Yang Zhang, Yue Wang, Zhihan Lu, Vivek Boominathan, Ashok Veeraraghavan, Yingyan Lin
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 |
|---|---|---|---|
| co-optimize the phase mask with the CNN via NAS instead of fixed Gabor mask | SACoD (6 masks) · CIFAR-100 | accuracy (%): +5.46% higher accuracy, 64.1% fewer FLOPs | Improved |
| co-search the backend architecture with the mask instead of fixed MobileNetV2 Co-train | SACoD (6 masks) · CIFAR-10 | accuracy (%): +0.01% accuracy, 44.1% fewer FLOPs | Improved |
| co-search the backend architecture with the mask instead of fixed Gabor mask | SACoD (6 masks) · CIFAR-10 | accuracy (%): +1.45% accuracy, 70.9% fewer FLOPs | Improved |
| use SACoD instead of Gabor-mask under extremely low FLOPs/energy | SACoD · FlatCam Face | accuracy (%): +8.75% higher accuracy | Improved |
| use SACoD instead of Gabor-mask under extremely low FLOPs/energy | SACoD · Head Pose | accuracy (%): +5.85% higher accuracy | Improved |
| use SACoD instead of Co-train on segmentation with 6 masks | SACoD · Cityscapes | mIOU: 68.8 -> 71.6 | Improved |
| use SACoD instead of Co-train on horse2zebra with 2 masks | SACoD · horse2zebra | FID: 66.82 -> 89.80 | Hurt |
| use SACoD instead of Gabor-mask on horse2zebra with 4 masks | SACoD · horse2zebra | FID: 106.27 -> 86.00 | Improved |
| use fabricated masks instead of simulated masks | SACoD · CIFAR-10 | accuracy (%): 94.41 -> 90.02 | Hurt |
| use SACoD masks instead of Gabor-mask in fabricated system | SACoD · CIFAR-10 (fabricated) | accuracy (%): 87.17 -> 90.02 | Improved |
| customize the mask per task instead of transferring masks from FlatCam Face (ablation) | SACoD · CIFAR-10 | accuracy (%): 93.10 -> 94.41 | Improved |
| customize the mask per task instead of transferring masks from FlatCam Face (ablation) | SACoD · CIFAR-100 | accuracy (%): 72.50 -> 76.67 | Improved |
Closest papers by meaning
- ASP Vision: Optically Computing the First Layer of Convolutional Neural Networks Using Angle Sensitive Pixels (CVPR 2016 · similarity 0.83)
- Reconfiguring the Imaging Pipeline for Computer Vision (ICCV 2017 · similarity 0.82)
- Constrained deep neural network architecture search for IoT devices accounting for hardware calibration (NeurIPS 2019 · similarity 0.82)
- SpArSe: Sparse Architecture Search for CNNs on Resource-Constrained Microcontrollers (NeurIPS 2019 · similarity 0.82)
- RepViT: Revisiting Mobile CNN From ViT Perspective (CVPR 2024 · similarity 0.80)
- Real-time Core-Periphery Guided ViT with Smart Data Layout Selection on Mobile Devices (NeurIPS 2024 · similarity 0.80)
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.
