Research map / Detection and segmentation / Open-set and few-shot detection
Fantastic Tractor-Dogs and How Not to Find Them With Open-Vocabulary Detectors
Frank Ruis, Gertjan J. Burghouts, Hugo Kuijf
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 |
|---|---|---|---|
| append attention sink tokens to the prompt | MM-GDINO Swin-T · LVIS MiniVal (negative-class prompts) | AP at FPRbg 0.01: 0.076 -> 0.214 | Improved |
| append attention sink tokens to the prompt | LLMDet Swin-T · LVIS MiniVal (negative-class prompts) | AP at FPRbg 0.01: 0.045 -> 0.223 | Improved |
| append attention sink tokens to the prompt | LLMDet Swin-T · LVIS MiniVal (negative-class prompts) | AP at FPRbg 0.05: 0.140 -> 0.326 | Improved |
| append attention sink tokens to the prompt | LLMDet Swin-B · LVIS MiniVal (negative-class prompts) | AP at FPRbg 0.05: 0.125 -> 0.349 | Improved |
| append attention sink tokens to the prompt | GLIP Swin-T · LVIS MiniVal (negative-class prompts) | standard AP (no FPR filter): 0.333 -> 0.302 | Hurt |
| append attention sink tokens to the prompt | LLMDet Swin-L · LVIS MiniVal (negative-class prompts) | standard AP (no FPR filter): 0.488 -> 0.502 | No clear effect |
| append attention sink tokens to the prompt | GDINO Swin-T · POPE detection FPRbg 0.05 | AP: 0.413 -> 0.413 | No clear effect |
| append attention sink tokens to the prompt | LLMDet-T · POPE detection FPRbg 0.05 | AP: 0.278 -> 0.417 | Improved |
| learn the attention sink embeddings by textual inversion instead of training-free init (ablation) | attention sinks · VOC-based sink training | AP at low FPRbg | No clear effect |
| disable problematic attention heads instead of using attention sinks (ablation) | early-fusion OVD · background false positives | AP at low FPRbg | Mixed |
| use more than about 8 attention sinks (24 optimal) (ablation) | LLMDet-T · LVIS MiniVal | AP at low FPRbg | Improved |
| add vision sinks to vision features in addition to language sinks (ablation) | attention sinks · LVIS MiniVal | AP at low FPRbg | No clear effect |
Closest papers by meaning
- Language-conditioned Detection Transformer (CVPR 2024 · similarity 0.87)
- Taming Self-Training for Open-Vocabulary Object Detection (CVPR 2024 · similarity 0.87)
- Learning Background Prompts to Discover Implicit Knowledge for Open Vocabulary Object Detection (CVPR 2024 · similarity 0.86)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 2025 · similarity 0.86)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- How to Evaluate the Generalization of Detection? A Benchmark for Comprehensive Open-Vocabulary Detection (AAAI 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.
