OWL-ViT
Google's open-vocabulary detector that finds objects described by free text, plus one-shot detection from a single example image.
About
OWL-ViT (Open-World Localization with Vision Transformers) is the Google Research model that made open-vocabulary detection practical. It takes a standard Vision Transformer pretrained on image-text pairs, drops the final token pooling layer, and attaches lightweight box and class prediction heads to every output token. At inference you hand it an image and a free-text query, and it returns boxes for whatever the text describes, including classes it never saw in training. It also does image-conditional detection: show it one example picture of an object and it finds visually similar instances.
Published at ECCV 2022, it reached 31.2% zero-shot LVIS APr with a ViT-L/14 backbone and pushed image-conditional COCO AP50 from 26.0 to 41.8 over prior work. Code and checkpoints ship under Apache 2.0, and the weights are on Hugging Face (google/owlvit-base-patch32 and google/owlvit-large-patch14), which made OWL-ViT the default baseline that later open-vocabulary detectors measured themselves against.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Author-reported (ECCV 2022): 31.2% zero-shot LVIS APr (ViT-L/14); image-conditional COCO AP50 41.8 vs 26.0 prior SOTA
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: ECCV 2022 paper (arXiv:2205.06230) with LVIS and COCO numbers confirmed via the official scenic repo README; Apache-2.0 license for code and checkpoints. Strengths: landmark open-vocab detector, Hugging Face checkpoints widely reused, simple architecture. Docked for: (1) repo lives inside the large scenic monorepo, harder to navigate than a standalone release; (2) high-resolution inference is slower than single-shot detectors like YOLO-World. License: Apache-2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · ECCV 2022
- · 31.2% zero-shot LVIS APr (ViT-L/14), author-reported
- · Hugging Face checkpoints google/owlvit-base-patch32 and google/owlvit-large-patch14