Research map / Detection and segmentation / Open-set and few-shot detection
Scaling Open-Vocabulary Object Detection
Matthias Minderer, Alexey A. Gritsenko, Neil Houlsby
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 |
|---|---|---|---|
| Self-train on WebLI image-text data using N-gram queries | OWL detector · LVIS val rare | AP: 31.2 to 34.9 | Improved |
| Fine-tune the self-trained model on LVIS base | OWL-ST CLIP L/14 · LVIS val rare | AP: 34.9 to 44.6 | Improved |
| Fine-tune the self-trained model on LVIS base | OWL-ST CLIP L/14 · ODinW 13 | mean AP: 53.0 to 50.1 | Hurt |
| Add curated-class queries to the N-gram pseudo-label vocabulary | OWL-ST+FT CLIP L/14 · LVIS val rare | AP: 44.6 to 45.9 | Improved |
| Use machine-generated N-gram queries rather than only a human-curated vocabulary | OWL-ST+FT CLIP L/14 · ODinW 13 | mean AP: 48.7 to 50.1 | Improved |
| Average the weights of the self-trained and fine-tuned checkpoints | OWL-ST CLIP L/14 · ODinW 13 | mean AP: 48.7 to 56.3 | Improved |
| Drop 0.50 of image tokens by pixel variance (ablation) | OWL-ViT L/14 · LVIS val rare | AP: 31.8 ± 1.16 to 30.8 | No clear effect |
| Drop 0.70 of image tokens by pixel variance (ablation) | OWL-ViT L/14 · LVIS val rare | AP: 31.8 ± 1.16 to 28.2 | Hurt |
| Use a strict pseudo-annotation confidence threshold instead of a moderate threshold (ablation) | OWL-ST · LVIS | AP | Hurt |
| Train on curated classes instead of diverse machine-generated text queries (ablation) | OWL-ST · unseen classes and ODinW | AP | Hurt |
| Co-train on WebLI and LVIS rather than average self-trained and fine-tuned weights (ablation) | OWL-ST · open-world performance | AP | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DetCLIPv2 — DetCLIP: Dictionary-Enriched Visual-Concept Paralleled Pre-training for Open-world Detection (NeurIPS 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Taming Self-Training for Open-Vocabulary Object Detection (CVPR 2024 · similarity 0.92)
- MarvelOVD: Marrying Object Recognition and Vision-Language Models for Robust Open-Vocabulary Object Detection (ECCV 2024 · similarity 0.91)
- How to Evaluate the Generalization of Detection? A Benchmark for Comprehensive Open-Vocabulary Detection (AAAI 2024 · similarity 0.89)
- Open-Vocabulary Object Detection via Language Hierarchy (NeurIPS 2024 · similarity 0.89)
- OW-OVD: Unified Open World and Open Vocabulary Object Detection (CVPR 2025 · similarity 0.88)
- Language-conditioned Detection Transformer (CVPR 2024 · similarity 0.88)
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.
