Research map / Label-efficient and robust learning / Recognition and long-tail
Studying Very Low Resolution Recognition Using Deep Networks
Zhangyang Wang, Shiyu Chang, Yingzhen Yang, Ding Liu, Thomas S. Huang
Check 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 |
|---|---|---|---|
| Add SR pre-training using HR training images | Basic CNN · CIFAR-10 | top-1 classification error (%): 27.47 -> 25.15; 2.32% | Improved |
| Add SR pre-training using HR training images | Basic CNN · CIFAR-100 | top-1 classification error (%): 3.45%; 46.50% | Improved |
| Add LR-HR feature transfer | Model II · CIFAR-10 and CIFAR-100 | top-1 classification error (%): 21.72%; 43.03%; 3.43%; 3.47% | Improved |
| Replace full LR-HR feature coupling with partial coupling | Model III · CIFAR-10 and CIFAR-100 | top-1 classification error (%): 18.77; 41.64; 2.95; 4.86 | Improved |
| Increase the third-layer coupled ratio from 0.75 to 1.00 while holding the first two at 0.50 (ablation) | Partially coupled network · CIFAR-10 | top-1 classification error (%): 19.91% -> 20.72% | Hurt |
| Use Huber loss instead of MSE in SR pre-training (ablation) | Model IV · CIFAR-10 and CIFAR-100, clean | top-1 classification error (%): 18.77 -> 18.81; 41.64 -> 41.70 | No clear effect |
| Use Huber loss instead of MSE in SR pre-training (ablation) | Model IV · CIFAR-10 and CIFAR-100 with 5%, 10%, and 15% corruption | top-1 classification error (%): 19.99 -> 19.04; 23.03 -> 20.28; 25.05 -> 22.24; 43.43 -> 42.18; 46.20 -> 43.79; 49.49 -> 46.18 | Improved |
| Add SR pre-training, LR-HR transfer, and partial coupling | VLRR face identification model · UCCS face identification | top-1 accuracy: 40.97%@top-1; around 10%@top-1 over the simplest baseline | Improved |
| Add Huber loss in pre-training (ablation) | Model IV · UCCS face identification | top-1 accuracy: 1.06%@top-1 | Improved |
| Add Huber loss in pre-training (ablation) | Model IV · SVHN | top-1 accuracy: 4.05%@top-1 | Improved |
| Add LR-HR transfer (ablation) | Model II · SVHN | top-1 accuracy: 4.03%@top-1 | Improved |
| Use Huber loss in pre-training (ablation) | Model IV · VFR font recognition | recognition error | No clear effect |
Closest papers by meaning
- Accurate Image Super-Resolution Using Very Deep Convolutional Networks (CVPR 2016 · similarity 0.86)
- Rethinking Resolution in the Context of Efficient Video Recognition (NeurIPS 2022 · similarity 0.86)
- Learning a Single Convolutional Super-Resolution Network for Multiple Degradations (CVPR 2018 · similarity 0.86)
- SAVSR: Arbitrary-Scale Video Super-Resolution via a Learned Scale-Adaptive Network (AAAI 2024 · similarity 0.86)
- Exploring the Low-Pass Filtering Behavior in Image Super-Resolution (ICML 2024 · similarity 0.85)
- Beyond Image Super-Resolution for Image Recognition with Task-Driven Perceptual Loss (CVPR 2024 · similarity 0.84)
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.
