Research map / Detection and segmentation / CNN features
Mining Mid-level Visual Patterns with Deep CNN Activations
Yao Li, Lingqiao Liu, Chunhua Shen, Anton van den Hengel
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 |
|---|---|---|---|
| encode with Bag-of-Elements instead of Bag-of-Patterns (ablation) | MDPM · Pascal VOC 2007 | mAP | Improved |
| add merging of mid-level elements before encoding (BoE-M vs BoE-S) (ablation) | MDPM BoE · Pascal VOC 2007 | mAP | Improved |
| add pattern merging instead of pattern selection only (ablation) | MDPM BoE (VGG-VD, 50 elements) · Pascal VOC 2007 | mAP (%): 86.2 -> 87.3 | Improved |
| use more mid-level elements (10 to 50) (ablation) | MDPM BoE · Pascal VOC 2007 | mAP | Improved |
| increase transaction length from 10 to 20 (ablation) | MDPM BoE (VGG-VD) · Pascal VOC 2007 | mAP (%): 85.4 -> 87.3 | Improved |
| increase transaction length from 20 to 30 (ablation) | MDPM BoE (VGG-VD) · Pascal VOC 2007 | mAP (%): 87.3 -> 87.6 | No clear effect |
| raise merging threshold Th from 50 to 150 (ablation) | MDPM BoE (VGG-VD) · Pascal VOC 2007 | mAP (%): 86.4 -> 87.3 | Improved |
| raise merging threshold Th from 150 to 200 (ablation) | MDPM BoE (VGG-VD) · Pascal VOC 2007 | mAP (%): 87.3 -> 87.0 | No clear effect |
| use BoE with VGG-VD instead of global FC feature | MDPM · Pascal VOC 2007 | mAP (%): 79.6 -> 87.3 | Improved |
| use BoE with VGG-VD instead of global FC feature | MDPM · MIT Indoor | accuracy (%): 68.87 -> 77.63 | Improved |
| replace pre-trained VGG-VD with fine-tuned VGG-VD for mining | MDPM · MIT Indoor | accuracy (%): 77.63 -> 71.82 | Hurt |
| replace LDA-Retrained detectors with pattern mining | MDPM · MIT Indoor | accuracy (%): 62.30 -> 69.69 | Improved |
Closest papers by meaning
- Neural Motifs: Scene Graph Parsing With Global Context (CVPR 2018 · similarity 0.87)
- DeepCAMP: Deep Convolutional Action & Attribute Mid-Level Patterns (CVPR 2016 · similarity 0.86)
- Bilinear Convolutional Neural Networks for Fine-Grained Visual Recognition (TPAMI 2017 · similarity 0.86)
- Automatic Attribute Discovery with Neural Activations (ECCV 2016 · similarity 0.86)
- Deep Co-Occurrence Feature Learning for Visual Object Recognition (CVPR 2017 · similarity 0.85)
- Learning Visual Features from Large Weakly Supervised Data (ECCV 2016 · 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.
