Research map / Label-efficient and robust learning / Domain adaptation and generalization
Semantic Clustering for Robust Fine-Grained Scene Recognition
Marian George, Mandar Dixit, Gábor Zogg, Nuno Vasconcelos
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 semantic clustering (latent topics with per-topic classifiers) (ablation) | OOM · SnapStore | accuracy (%): 45.7 -> 47.9 | Improved |
| add semantic clustering (ablation) | OOM (soft detections) · MIT67 | accuracy (%): 68.2 -> 68.6 | No clear effect |
| add semantic clustering (ablation) | OOM · cross-domain average (7 scenarios) | accuracy (%): 58.5 -> 60.4 | Improved |
| add semantic clustering (ablation) | ObjectBank · cross-domain average (7 scenarios) | accuracy (%): 32.9 -> 33.4 | No clear effect |
| cluster DeCaF features with k-means (ablation) | DeCaF · cross-domain average (7 scenarios) | accuracy (%): 58.8 -> 56.7 | Hurt |
| replace raw detection scores with OOM scene probabilities | ObjectBank (CNN detections) · SnapStore | accuracy (%): 34.8 -> 45.4 | Improved |
| use discriminant object selection (fewer objects) (ablation) | OOM (soft detections) · SnapStore | accuracy (%): 43.1 -> 45.4 | Improved |
| use discriminant object selection (fewer objects) (ablation) | OOM (soft detections) · MIT67 | accuracy (%): 68.0 -> 68.2 -> 66.4 | Mixed |
| fine-tune Places CNN instead of using fc7 features | Places CNN · SnapStore | accuracy (%): 44.2 -> 42.4 | Hurt |
| use max pooling instead of average pooling over topic classifiers (ablation) | semantic clustering · SnapStore | accuracy (%) | Hurt |
| vary number of k-means clusters k (ablation) | semantic clustering · SnapStore | accuracy (%) | No clear effect |
| use hard (RCNN, 200 objects) instead of soft detections | OOM · MIT67 | accuracy (%): 68.2 -> 49.4 | Hurt |
Closest papers by meaning
- Knowledge Mining With Scene Text for Fine-Grained Recognition (CVPR 2022 · similarity 0.88)
- Fine-Grained Recognition in the Wild: A Multi-Task Domain Adaptation Approach (ICCV 2017 · similarity 0.88)
- Democratizing Fine-grained Visual Recognition with Large Language Models (ICLR 2024 · similarity 0.87)
- Fine-Grained Perception in Panoramic Scenes: A Novel Task, Dataset, and Method for Object Importance Ranking (AAAI 2025 · similarity 0.87)
- The Devil is in the Fine-Grained Details: Evaluating Open-Vocabulary Object Detectors for Fine-Grained Understanding (CVPR 2024 · similarity 0.87)
- GUIDED: Granular Understanding via Identification, Detection, and Discrimination for Fine-Grained Open-Vocabulary Object Detection (NeurIPS 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.
