Research map / Architectures and efficiency / CNN design
An Empirical Evaluation of Current Convolutional Architectures' Ability to Manage Nuisance Location and Scale Variability
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a 10-pixel rim around the ground-truth object box (ablation) | AlexNet · ImageNet 2014 validation | top-5 error: 20.41 -> 17.66 | Helped |
| Add a 10-pixel rim around the ground-truth object box (ablation) | VGG16 · ImageNet 2014 validation | top-5 error: 12.44 -> 10.91 | Helped |
| Average 8 domain sizes around the ground-truth box instead of one padded box (ablation) | AlexNet · ImageNet 2014 validation | top-5 error: 17.66 -> 14.43 | Helped |
| Average 8 domain sizes around the ground-truth box instead of one padded box (ablation) | VGG16 · ImageNet 2014 validation | top-5 error: 10.91 -> 8.66 | Helped |
| Use 50 regular crops at 3 scales rather than one whole-image input (ablation) | AlexNet · ImageNet 2014 validation | top-5 error: 19.96 -> 17.55 | Helped |
| Use 50 regular crops at 3 scales rather than one whole-image input (ablation) | VGG16 · ImageNet 2014 validation | top-5 error: 13.24 -> 8.85 | Helped |
| Add 10 concentric domain sizes to whole-image inference (ablation) | AlexNet · ImageNet 2014 validation | top-5 error: 19.96 -> 17.86 | Helped |
| Add 10 concentric domain sizes to whole-image inference (ablation) | VGG16 · ImageNet 2014 validation | top-5 error: 13.24 -> 9.46 | Helped |
| Combine 10 regular crops, 10 domain sizes, and 12 entropy-selected proposals | AlexNet · ImageNet 2014 validation | top-5 error: 17.55 -> 15.83 | Helped |
| Combine 10 regular crops, 10 domain sizes, and 12 entropy-selected proposals | VGG16 · ImageNet 2014 validation | top-5 error: 8.85 -> 8.01 | Helped |
| Weight selected posteriors by inverse entropy rather than uniformly (ablation) | AlexNet proposed sampler · ImageNet 2014 validation | top-5 error: 15.83 -> 15.82 | Helped |
| Weight selected posteriors by inverse entropy rather than uniformly (ablation) | VGG16 proposed sampler · ImageNet 2014 validation | top-5 error: 8.01 -> 8.02 | No clear effect |
Closest papers by meaning
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.90)
- Contextual Convolutional Networks (ICLR 2023 · similarity 0.89)
- Refining Architectures of Deep Convolutional Neural Networks (CVPR 2016 · similarity 0.89)
- Involution: Inverting the Inherence of Convolution for Visual Recognition (CVPR 2021 · similarity 0.88)
- Glance and Focus: a Dynamic Approach to Reducing Spatial Redundancy in Image Classification (NeurIPS 2020 · similarity 0.88)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.87)
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.