Grounding DINO

A local AI object detection model finds objects from text descriptions and returns bounding boxes, with PyTorch support for CPU or CUDA hardware.

Grounding DINO locates objects in images from category names or descriptive text. It's for developers and computer vision researchers building detection or annotation workflows that need labels beyond a fixed category list.

Text-directed detection: Supply an image and a description to find matching objects without training a separate detector for each requested category.
Bounding boxes and labels: Get object locations with text similarity scores, and focus detection on a specific phrase within a longer description.
Segmentation and tracking integrations: Grounded-SAM pairs the detector with Segment Anything for segmentation, while the separate Grounded SAM 2 project combines it with SAM 2 for object tracking.
Image editing integrations: Example notebooks connect detected objects to Stable Diffusion and GLIGEN to guide image edits.

The open-source Python implementation uses PyTorch and is licensed under Apache 2.0. Pretrained models include GroundingDINO-T and GroundingDINO-B. Detection can run on your own hardware in CPU-only mode or use CUDA with an NVIDIA GPU.

A Gradio demo provides a browser interface to the model. Hugging Face and Colab demos offer hosted ways to try detection outside your machine.

Computer Vision Models Vision-Language Models Docker Web Hugging Face integration Multimodal input