TinyML computer vision using coarsely-quantized log-gradient input images
Abstract/Contents
- Abstract
- This thesis studies the merits of applying log-gradient input images to convolutional neural networks (CNNs) for tinyML computer vision (CV). We show that log gradients enable: (i) aggressive 1-bit quantization of first-layer inputs, (ii) potential CNN resource reductions, (iii) inherent insensitivity to illumination changes (1.7% accuracy loss across 2^{-5}...2^3 brightness variation vs. up to 10% for JPEG), (iv) robustness to adversarial attacks (> 10% higher accuracy than JPEG-trained models), and (v) and insensitivity thresholds for quantization that confirms hardware implementability. We establish these results using the PASCAL RAW image data set and through a combination of experiments using quantization threshold search, neural architecture search, and a fixed three-layer network. The latter reveal that training on log-gradient images leads to higher filter similarity, making the CNN more prunable. The combined benefits of aggressive first-layer quantization, CNN resource reductions, and operation without tight exposure control and image signal processing (ISP) are helpful for pushing tinyML CV toward its ultimate efficiency limits. Dataset related issues currently prevailing in machine-centered vision tasks are reviewed and analyzed, with a focus on RAW image datasets. By approximating RAW data, our proposed method shows potential to generalize to larger datasets. Taken together, we demonstrate an energy efficient and robust tinyML computer vision system approach using coarsely-quantized log-gradient input images.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2023; ©2023 |
| Publication date | 2023; 2023 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Lu, Qianyun |
|---|---|
| Degree supervisor | Murmann, Boris |
| Thesis advisor | Murmann, Boris |
| Thesis advisor | Pilanci, Mert |
| Thesis advisor | Wetzstein, Gordon |
| Degree committee member | Pilanci, Mert |
| Degree committee member | Wetzstein, Gordon |
| Associated with | Stanford University, School of Engineering |
| Associated with | Stanford University, Department of Electrical Engineering |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Qianyun Lu. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Thesis Ph.D. Stanford University 2023. |
| Location | https://purl.stanford.edu/fb372vt6975 |
Access conditions
- Copyright
- © 2023 by Qianyun Lu
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...