Research map / Detection and segmentation / CNN features
SPDA-CNN: Unifying Semantic Part Detection and Abstraction for Fine-Grained Recognition
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 |
|---|---|---|---|
| replace bottom-up proposals with top-down k-NN part proposals | CUB-2011 part detection · CUB-2011 | mAP: 63.12% -> 72.72% | Helped |
| increase k from 1 to 20 in k-NN proposals (ablation) | SPDA-CNN part proposals · CUB-2011 | mAP: 52.83% -> 72.72% | Helped |
| use weak rather than strong prior for k-NN proposals (ablation) | SPDA-CNN part proposals · CUB-2011 | mAP: 71.79% -> 72.72% | Helped |
| add semantic part layers instead of using object-only features (ablation) | CLS-NET · CUB-2011 classification with oracle parts | accuracy: 67.02 -> 79.46 | Helped |
| use 7 parts instead of 2 (ablation) | CLS-NET · CUB-2011 classification with oracle parts | accuracy: 77.08 -> 79.46 | Helped |
| use 1 pfc layer rather than 2 (ablation) | 7-part CLS-NET · CUB-2011 classification with oracle parts | accuracy: 79.10 vs 79.46 | No clear effect |
| add whole-object features to the 7-part model (ablation) | CLS-NET · CUB-2011 classification with oracle parts | accuracy: 79.46 -> 78.17 | Hurt |
| ensemble CLS-NET variants | CLS-NET · CUB-2011 classification with oracle parts | accuracy: 79.46 -> 81.96 | Helped |
| fine-tune the classifier using detected training-image parts | SPDA-CNN · CUB-2011 classification | accuracy: 78.15% -> 78.93% | Helped |
| use detected rather than oracle part annotations for classification | SPDA-CNN · CUB-2011 classification | accuracy: 79.46 -> 78.15% | Hurt |
| use detected parts with SPDA-CNN rather than the strongest reported VGGNet baseline | SPDA-CNN · CUB-2011 classification | accuracy: 84.10 -> 85.14 | Helped |
| add head-part features rather than use object or other-part features | pfc features for attribute prediction · CUB-2011 head attribute prediction | AUC: mean 0.69 -> 0.73 | Helped |
Closest papers by meaning
- Part-Stacked CNN for Fine-Grained Visual Categorization (CVPR 2016 · similarity 0.91)
- Interpretable and Accurate Fine-grained Recognition via Region Grouping (CVPR 2020 · similarity 0.90)
- Weakly Supervised Learning of Part Selection Model with Spatial Constraints for Fine-Grained Image Classification (AAAI 2017 · similarity 0.89)
- Beyond Holistic Object Recognition: Enriching Image Understanding With Part States (CVPR 2018 · similarity 0.89)
- Deep Co-Occurrence Feature Learning for Visual Object Recognition (CVPR 2017 · similarity 0.88)
- DoGA: Enhancing Grounded Object Detection via Grouped Pre-Training with Attributes (AAAI 2025 · 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.