rex-omni-torch#

Vision-language detector that finds any object you name by predicting quantized coordinate tokens.

Details

  • Model name: rex-omni-torch

  • Model source: https://huggingface.co/IDEA-Research/Rex-Omni

  • Model author: IDEA Research

  • Model license: IDEA License 1.0 (research-only); based on Qwen2.5-VL under the Qwen RESEARCH LICENSE

  • Model size: 7.57 GB

  • Exposes embeddings? no

  • Tags: detection, vlm, torch, transformer, zero-shot

Requirements

  • Packages: torch, torchvision, transformers>=4.49,<5, accelerate

  • CPU support

    • yes

  • GPU support

    • yes

Example usage

 1import fiftyone as fo
 2import fiftyone.zoo as foz
 3
 4dataset = foz.load_zoo_dataset(
 5    "coco-2017",
 6    split="validation",
 7    dataset_name=fo.get_default_dataset_name(),
 8    max_samples=50,
 9    shuffle=True,
10)
11
12model = foz.load_zoo_model("rex-omni-torch")
13
14dataset.apply_model(model, label_field="predictions")
15
16session = fo.launch_app(dataset)