OWLViT

OWL-ViT extends CLIP-based vision–language models to perform open-vocabulary object detection by aligning image regions with textual descriptions, enabling zero-shot detection without task-specific training.

Original paper: Simple Open-Vocabulary Object Detection with Vision Transformers

OWLv1 CLIP ViT-B/32

This model uses the OWL-ViT v1 with a CLIP ViT-B/32 Transformer architecture as an image encoder and a masked self-attention Transformer as a text encoder, leveraging the vision–language alignment of CLIP to detect objects specified by arbitrary text queries. It is well suited for applications such as open-vocabulary detection, image search, and real-time visual understanding across diverse domains.

Model Configuration:

• Reference implementation: OWLv1 CLIP ViT-B/32
• Original Weight: owlvit-base-patch32
• Resolution: 3x768x768
• Support Cooper version:
  • Cooper SDK: [2.5.4]
  • Cooper Foundry: [2.3]