Research map / Detection and segmentation / CNN features
Learning Deep Features for Discriminative Localization
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 |
|---|---|---|---|
| Use global average pooling instead of global max pooling (ablation) | GoogLeNet localization · ILSVRC 2014 val | top-5 localization error: 45.26 -> 43.00 | Helped |
| Use global average pooling instead of global max pooling (ablation) | GoogLeNet classification · ILSVRC 2014 val | top-5 classification error: 13.9 -> 13.2 | No clear effect |
| Use CAM instead of backpropagation for localization | GoogLeNet · ILSVRC 2014 val | top-5 localization error: 50.55 -> 43.00 | Helped |
| Use higher mapping resolution with the GAP architecture | GoogLeNet localization · ILSVRC 2014 val | top-5 localization error: 49.34 -> 43.00 | Helped |
| Use multiple heuristic bounding boxes from predicted classes (ablation) | GoogLeNet-GAP · ILSVRC 2014 test | top-5 localization error: 42.9 -> 37.1 | Helped |
| Extract features from CAM-localized crops instead of full images (ablation) | GoogLeNet-GAP · CUB-200-2011 | accuracy: 63.0% -> 67.8% | Helped |
| Use annotated bounding-box crops instead of full images | GoogLeNet-GAP · CUB-200-2011 | accuracy: 63.0% -> 70.5% | Helped |
| Replace the original AlexNet head with GAP (ablation) | AlexNet · ILSVRC 2014 val | top-5 classification error: 19.5 -> 26.3 | Hurt |
| Add two convolutional layers before GAP (ablation) | AlexNet-GAP · ILSVRC 2014 val | top-5 classification error: 26.3 -> 20.9 | Helped |
Closest papers by meaning
- Kernelized Subspace Pooling for Deep Local Descriptors (CVPR 2018 · similarity 0.88)
- Revisiting Dilated Convolution: A Simple Approach for Weakly- and Semi-Supervised Semantic Segmentation (CVPR 2018 · similarity 0.87)
- Local Features and Visual Words Emerge in Activations (CVPR 2019 · similarity 0.86)
- What Deep CNNs Benefit From Global Covariance Pooling: An Optimization Perspective (CVPR 2020 · similarity 0.86)
- Recognize Any Regions (NeurIPS 2024 · similarity 0.86)
- CAM Back Again: Large Kernel CNNs from a Weakly Supervised Object Localization Perspective (CVPR 2024 · similarity 0.85)
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.