Research map / Vision-language and multimodal / Human and embodied modeling
Concept Drift and Long-Tailed Distribution in Fine-Grained Visual Categorization: Benchmark and Method
Shuo Ye, Shiming Chen, Ruxin Wang, Tianxu Wu, Salman Khan, Fahad Shahbaz Khan, Ling Shao
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 |
|---|---|---|---|
| train on all data instead of ignoring concept-drift data | NTS-Net (ResNet-50) · CDLT-cd | top-1 accuracy: 69.00 -> 74.32 | Improved |
| train on all data instead of ignoring concept-drift data | ViT-B-16 IMS · CDLT-cd | top-1 accuracy: 70.82 -> 77.92 | Improved |
| use range loss instead of softmax loss | ResNet-50 · CDLT long-tailed overall | top-1 accuracy: 62.27 -> 72.30 | Improved |
| use under-sampling instead of softmax loss | ResNet-50 · CDLT long-tailed overall | top-1 accuracy: 62.27 -> 46.21 | Hurt |
| use over-sampling instead of softmax loss | ResNet-50 · CDLT long-tailed overall | top-1 accuracy: 62.27 -> 62.20 | No clear effect |
| use focal loss instead of softmax loss | ResNet-50 · CDLT long-tailed overall | top-1 accuracy: 62.27 -> 66.31 | Improved |
| use class-balanced loss instead of softmax loss | ResNet-50 · CDLT long-tailed overall | top-1 accuracy: 62.27 -> 67.45 | Improved |
| apply frequency-based feature recombination (diffusion low-frequency generation and high-frequency shuffling) | ResNet-50 · CDLT long-tailed overall | top-1 accuracy: 72.30 -> 75.18 | Improved |
| apply feature recombination instead of NTS-Net | ResNet-50 · CDLT | top-1 accuracy: 78.20 -> 78.82 | Improved |
| fine-tune CLIP with a text-encoder adapter | CLIP ViT-B · CDLT | accuracy: 3.03 -> 2.02 | No clear effect |
| fine-tune CLIP with a text-encoder adapter | CLIP ViT-B · CUB-200-2011 | accuracy: 50.89 -> 58.91 | Improved |
| fine-tune CLIP with a text-encoder adapter | CLIP ViT-B · Stanford Cars | accuracy: 57.73 -> 53.22 | Hurt |
Closest papers by meaning
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.87)
- Advancing Fine-Grained Classification by Structure and Subject Preserving Augmentation (NeurIPS 2024 · similarity 0.87)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.86)
- SELECT: A Large-Scale Benchmark of Data Curation Strategies for Image Classification (NeurIPS 2024 · similarity 0.86)
- Benchmarking Large Vision-Language Models on Fine-Grained Image Tasks: A Comprehensive Evaluation (ICLR 2026 · similarity 0.86)
- DiffuLT: Diffusion for Long-tail Recognition Without External Knowledge (NeurIPS 2024 · 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.
