Research map / Label-efficient and robust learning / Recognition and long-tail
Effects of Image Degradation and Degradation Removal to CNN-Based Image Classification
Yanting Pei, Yaping Huang, Qi Zou, Xingyuan Zhang, Song Wang
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 |
|---|---|---|---|
| train on clear images instead of matched degraded images | VGGNet-16 · Haze beta=5 test | accuracy (%): 67.8 -> 33.0 | Hurt |
| train on clear images instead of matched motion-blurred images | VGGNet-16 · Motion blur (40,40) test | accuracy (%): 59.0 -> 11.2 | Hurt |
| increase motion blur from (0,0) to (40,40) with matched training | VGGNet-16 · Motion blur | accuracy (%): 81.0 -> 59.0 | Hurt |
| use VGGNet-16 instead of AlexNet | CNN classifier · Caltech-256 clear | accuracy (%): 71.7 -> 81.0 | Improved |
| mix all degradation levels in training | VGGNet-16 · Haze beta=5 test | accuracy (%): 67.8 -> 69.2 | Improved |
| mix degraded levels into training, evaluated on clear images | VGGNet-16 · Caltech-256 clear test (hazy mix) | accuracy (%): 81.0 -> 79.7 | Hurt |
| add dehazing pre-processing at test time (train clear) | VGGNet-16 · Haze beta=5 test | accuracy (%): 33.0 -> 43.0 | Improved |
| add deblurring pre-processing at test time (train clear) | VGGNet-16 · Motion blur (15,15) test | accuracy (%): 45.1 -> 50.3 | Improved |
| add distortion correction at test time (train clear) | VGGNet-16 · Fish-eye e=1.75 test | accuracy (%): 25.4 -> 20.0 | Hurt |
| train on real hazy images instead of clear images | VGGNet-16 · Haze-20 test | accuracy (%): 62.3 -> 81.2 | Improved |
| combine hazy and clear images in training | VGGNet-16 · HazeClear-20 test | accuracy (%): 88.7 -> 97.7 | Improved |
Closest papers by meaning
- Deep Degradation Prior for Low-Quality Image Classification (CVPR 2020 · similarity 0.89)
- Deep Image Deblurring: A Survey (IJCV 2022 · similarity 0.87)
- Learning Degradation Representations for Image Deblurring (ECCV 2022 · similarity 0.86)
- Does Haze Removal Help CNN-based Image Classification? (ECCV 2018 · similarity 0.86)
- DDR: Exploiting Deep Degradation Response as Flexible Image Descriptor (NeurIPS 2024 · similarity 0.86)
- Universal Denoising Networks : A Novel CNN Architecture for Image Denoising (CVPR 2018 · similarity 0.86)
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.
