Research map / Detection and segmentation / Weakly supervised detection and segmentation
Top-Down Neural Attention by Excitation Backprop
Jianming Zhang, Zhe Lin, Jonathan Brandt, Xiaohui Shen, Stan Sclaroff
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 |
|---|---|---|---|
| add contrastive attention (c-MWP vs MWP) | Excitation Backprop · VOC07 Test (difficult subset), GoogleNet | Pointing Game accuracy (%): 60.4 -> 72.3 | Improved |
| add contrastive attention (c-MWP vs MWP) | Excitation Backprop · COCO Val (difficult subset), GoogleNet | Pointing Game accuracy (%): 37.1 -> 48.3 | Improved |
| add the probabilistic normalization factor Z_i (ablation) | contrastive Excitation Backprop · Pointing Game | Mean accuracy (%): 41.6 -> 70.6 | Improved |
| propagate contrastive signal in a single pass instead of subtracting after L1 normalization (ablation) | contrastive Excitation Backprop · Pointing Game | Mean accuracy (%): 58.1 -> 70.6 | Improved |
| apply contrastive attention to Deconv | Deconv · Pointing Game | Mean accuracy (%): 49.4 -> 67.7 | Improved |
| apply contrastive attention to CAM | CAM · Pointing Game | Mean accuracy (%): 61.9 -> 61.9 | No clear effect |
| apply contrastive attention to Grad | Grad · Pointing Game | Mean accuracy (%) | Hurt |
| use contrastive attention (c-MWP vs MWP) | Excitation Backprop · ImageNet Val dominant object localization | Localization error (%): 38.7 -> 57.0 | Hurt |
| use attention at I5b/out layer instead of intermediate layers (ablation) | c-MWP vs Grad/Deconv · Pointing Game, GoogleNet layer selection | Pointing Game accuracy | Mixed |
| add contrastive attention (c-MWP vs MWP) | Excitation Backprop · Flickr30k Entities, MCG | Recall@1 (%): 28.5 -> 26.2 | Hurt |
| add contrastive attention (c-MWP vs MWP) | Excitation Backprop · Flickr30k Entities, MCG | mAP (Group): 11.8 -> 15.2 | Improved |
| re-score MCG proposals with MWP attention | MCG_base · Flickr30k Entities | Recall@1 (%): 10.7 -> 28.5 | Improved |
Closest papers by meaning
- Guided Attention Inference Network (TPAMI 2019 · similarity 0.88)
- TDAM: Top-Down Attention Module for Contextually Guided Feature Selection in CNNs (ECCV 2022 · similarity 0.88)
- Tell Me Where to Look: Guided Attention Inference Network (CVPR 2018 · similarity 0.88)
- Selective Attention: Enhancing Transformer through Principled Context Control (NeurIPS 2024 · similarity 0.85)
- SAS: Simple Attention Sparsification via End-to-End Optimization of Context Ranking (NeurIPS 2026 · similarity 0.84)
- Attention layers provably solve single-location regression (ICLR 2025 · 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.
