Research map / Detection and segmentation / Open-set and few-shot detection
Compositional Convolutional Neural Networks: A Deep Architecture With Innate Robustness to Partial Occlusion
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 the VGG-16 fully connected head with a multi-layer generative compositional head | VGG-16 · PASCAL3D+ Vehicles, occlusion levels L0–L3 (mean) | classification accuracy: 83.6 -> 95.4 | Helped |
| combine pool4 and pool5 compositional features (ablation) | CompNet · PASCAL3D+ Vehicles, occlusion levels L0–L3 (mean) | classification accuracy: 93.0 -> 95.4 | Helped |
| use pool4 instead of pool5 features (ablation) | CompNet · PASCAL3D+ Vehicles, strong occlusion by natural objects (L2–L3) | classification accuracy: L2: 88.1 -> 92.5; L3: 72.8 -> 88.1 | Mixed |
| fine-tune on non-occluded MS-COCO images (ablation) | CompNet-Multi · MS-COCO Vehicles, real occlusions | classification accuracy: 89.9 -> 93.0 | Helped |
| add CutOut augmentation (ablation) | CompNet-Multi · MS-COCO Vehicles, real occlusions | classification accuracy: 93.0 -> 92.9 | No clear effect |
| add CutPaste augmentation (ablation) | CompNet-Multi · MS-COCO Vehicles, real occlusions | classification accuracy: 93.0 -> 94.4 | Helped |
| use pool5 instead of the multi-layer combination (ablation) | CompNet · MS-COCO Vehicles, CutPaste training | classification accuracy: 94.4 -> 93.5 | Hurt |
| augment VGG training with CutPaste (ablation) | VGG-16 · MS-COCO Vehicles, real occlusions | classification accuracy: 82.4 -> 90.6 | Helped |
| use a compositional head rather than TDAPNet | classification model · MS-COCO Vehicles, test on real occlusions, train on MS-COCO L0 (Avg) | classification accuracy: 86.5 -> 93.0 | Helped |
| use CompositionalNet rather than dictionary-based compositional models for occluder localization | occluder localization model · Occluded-Vehicle, false acceptance rate 0.2 | localization performance: 12% (white), 19% (noise), 6% (texture), 8% (objects) | Helped |
| train TDAPNet with non-occluded MS-COCO images (ablation) | TDAPNet · MS-COCO Vehicles, real occlusions | classification accuracy | Mixed |
Closest papers by meaning
- Robust Object Detection Under Occlusion With Context-Aware CompositionalNets (CVPR 2020 · similarity 0.93)
- Teaching Compositionality to CNNs (CVPR 2017 · similarity 0.90)
- A Bayesian Approach to OOD Robustness in Image Classification (CVPR 2024 · similarity 0.88)
- Not Only Text: Exploring Compositionality of Visual Representations in Vision-Language Models (CVPR 2025 · similarity 0.86)
- Learning to Compose: Improving Object Centric Learning by Injecting Compositionality (ICLR 2024 · similarity 0.85)
- ImageNet-trained CNNs are not biased towards texture: Revisiting feature reliance through controlled suppression (NeurIPS 2025 · 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.