Research map / Detection and segmentation / Open-set and few-shot detection
Do Semantic Parts Emerge in Convolutional Neural Networks?
Abel González-García, Davide Modolo, Vittorio Ferrari
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 bounding-box regression to filter stimulus detections (ablation) | best single filter, AlexNet-Object layer 5 · PASCAL-Part train | part detection mAP: 6.1 -> 19.0 | Improved |
| add GA filter combination instead of the single best filter (ablation) | AlexNet-Object layer 5 · PASCAL-Part train | part detection mAP: +5.2 mAP (19.0 -> 24.2) | Improved |
| add GA filter combination in layer 1 (ablation) | AlexNet-Object layer 1 · PASCAL-Part train | part detection mAP: +0.6 mAP only | No clear effect |
| select filters with GA instead of TopFilters (ablation) | AlexNet-Object layer 5 · PASCAL-Part train | part detection mAP: 18.8 -> 24.2 | Improved |
| use all filters simultaneously instead of GA combination (ablation) | AlexNet-Object layer 5 · PASCAL-Part train | part detection mAP: 24.2 -> 9.2 | Hurt |
| train for scene recognition (Places) instead of object detection | AlexNet · PASCAL-Part train | part detection mAP: -6.1% mAP compared to AlexNet-Object (24.2 -> 18.1) | Hurt |
| add ILSVRC12 pre-training before object detection training (ablation) | AlexNet-Object · PASCAL-Part train layer 5 | part detection mAP: 18.2 -> 24.2 | Improved |
| fine-tune for object detection on VOC12 instead of image classification only | AlexNet · PASCAL-Part train layer 5 | part detection mAP: 23.2 -> 24.2 | Improved |
| use deeper VGG16 instead of AlexNet | object detection finetuned network · PASCAL-Part train layer 5 | part detection mAP: 24.2 -> 26.1 | Improved |
| fine-tune VGG16 for detection instead of classification only | VGG16 · PASCAL-Part train layer 5_3 | part detection mAP: 22.1 -> 26.1 | Improved |
| use higher network layer | AlexNet-Object GA · PASCAL-Part train | part detection mAP: 9.6 -> 14.4 -> 19.1 -> 21.1 -> 24.2 | Improved |
| larger average part size | AlexNet-Object layer 5 GA · PASCAL-Part train | part AP correlation: Pearson correlation 0.7 | Improved |
Closest papers by meaning
- Growing Interpretable Part Graphs on ConvNets via Multi-Shot Learning (AAAI 2017 · similarity 0.90)
- SPDA-CNN: Unifying Semantic Part Detection and Abstraction for Fine-Grained Recognition (CVPR 2016 · similarity 0.89)
- Beyond Holistic Object Recognition: Enriching Image Understanding With Part States (CVPR 2018 · similarity 0.88)
- P-CNN: Part-Based Convolutional Neural Networks for Fine-Grained Visual Categorization (TPAMI 2019 · similarity 0.88)
- Deep Co-Occurrence Feature Learning for Visual Object Recognition (CVPR 2017 · similarity 0.87)
- Shape or Texture: Understanding Discriminative Features in CNNs (ICLR 2021 · 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.
