Research map / Image generation / Autoregressive and latent generation
Learning To Resize Images for Computer Vision Tasks
Hossein Talebi, Peyman Milanfar
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 |
|---|---|---|---|
| replace bilinear resizer with jointly trained learned resizer | Inception-v2 · ImageNet val | top-1 error: 26.7% -> 24.0% | Improved |
| replace bilinear resizer with jointly trained learned resizer | DenseNet-121 · ImageNet val | top-1 error: 33.1% -> 29.8% | Improved |
| replace bilinear resizer with jointly trained learned resizer | ResNet-50 · ImageNet val | top-1 error: 24.7% -> 23.0% | Improved |
| replace bilinear resizer with jointly trained learned resizer | MobileNet-v2 · ImageNet val | top-1 error: 29.5% -> 28.4% | Improved |
| replace bicubic resizer with learned resizer | Inception-v2 · AVA IQA | PLCC: 0.662 -> 0.686 | Improved |
| replace bicubic resizer with learned resizer | EfficientNet-b0 · AVA IQA | PLCC: 0.642 -> 0.671 | Improved |
| raise bilinear baseline input resolution to 448x448 instead of using learned resizer | MobileNet-v2 · ImageNet val | top-1 error: 28.4% -> 26.1% | Hurt |
| add larger resizer (r=2, n=16) to default resizer (ablation) | Inception-v2 · ImageNet val | top-1 error: 24.5% -> 25.5% | Hurt |
| add larger resizer (r=2, n=32) (ablation) | MobileNet-v2 · ImageNet val | top-1 error: 28.4% -> 28.3% | Improved |
| fine-tune resizer initialized from another model's resizer | DenseNet-121 · ImageNet val | top-1 error: 30.2% -> 29.7% | Improved |
| apply ImageNet-trained resizer to CIFAR-10 classifier | ResNet-50 · CIFAR-10 | top-1 error: 6.9% -> 0.4% lower | Improved |
| use learned resizer as pre-processor instead of prior enhancement method | ResNet-50 · ImageNet 4x down-scaled | top-1 error: 31.8% -> 31.1% | Improved |
Closest papers by meaning
- Revisiting ResNets: Improved Training and Scaling Strategies (NeurIPS 2021 · similarity 0.89)
- MULLER: Multilayer Laplacian Resizer for Vision (ICCV 2023 · similarity 0.89)
- Task-Aware Image Downscaling (ECCV 2018 · similarity 0.87)
- Shape Adaptor: A Learnable Resizing Module (ECCV 2020 · similarity 0.86)
- ResFormer: Scaling ViTs With Multi-Resolution Training (CVPR 2023 · similarity 0.86)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 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.
