Grounding DINO -- tiny (Swin-T)
open-vocabulary object detection, trained on Objects365, GoldG and Cap4M (OGC)
48.4 box AP on COCO, zero-shot, as reported by the authors

Copyright 2023 - present, IDEA Research.
Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang,
Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, Lei Zhang.

Source:  https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth
Project: https://github.com/IDEA-Research/GroundingDINO
Paper:   Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set
         Object Detection (ECCV 2024), arXiv:2303.05499
Licence: Apache-2.0 (full text in the LICENSE file beside this one)

Where that licence is claimed. The GitHub repository's LICENSE covers the code and its
README states no terms for the checkpoints. The authors publish these same files under
an explicit apache-2.0 tag here:
         https://huggingface.co/ShilongLiu/GroundingDINO
         https://huggingface.co/IDEA-Research/grounding-dino-tiny

The checkpoint carries a fine-tuned BERT-base text encoder under its `bert.*` keys. The
vocabulary that tokenizes prompts for it is Google's `bert-base-uncased` vocab.txt,
Apache-2.0, and ships inside mozo rather than being downloaded.