grounding_dino_tiny

Grounding DINO performs open-set, text-grounded object detection: a Swin backbone and a BERT text encoder feed a cross-modality transformer that grounds free-form text prompts to boxes.

Pure-Keras 3 model in kerasformers. The repo is self-describing: kf_config.json holds the architecture and tokenizer.json the text tokenizer, so it loads by its Hub repo id (weights download automatically):

from kerasformers.models.grounding_dino import (

    GroundingDinoDetect,

    GroundingDinoProcessor,

)

model = GroundingDinoDetect.from_weights("kerasformers/grounding_dino_tiny")

processor = GroundingDinoProcessor.from_weights("kerasformers/grounding_dino_tiny")

inputs = processor(images=image, text="a cat. a remote control.")

outputs = model(inputs)  # logits + pred_boxes