# fc-clip-coco-panoptic-torch Open-vocabulary panoptic segmentation with FC-CLIP (COCO); returns both thing and stuff segments as instance masks. Uses frozen ConvNeXt-Large CLIP backbone + Mask2Former decoder.. **Details** - Model name: `fc-clip-coco-panoptic-torch` - Model source: [https://huggingface.co/Voxel51/fc-clip](https://huggingface.co/Voxel51/fc-clip) - Model author: Jiarui Xu, et al. (ByteDance) - Model license: Apache 2.0 - Model size: 79.16 MB - Exposes embeddings? no - Tags: `panoptic, zero-shot, torch, transformers, official` **Requirements** - Packages: `torch, torchvision, transformers, open_clip_torch, safetensors, timm>=1.0.17` - CPU support - yes - GPU support - yes **Example usage** ```python import fiftyone as fo import fiftyone.zoo as foz dataset = foz.load_zoo_dataset( "coco-2017", split="validation", dataset_name=fo.get_default_dataset_name(), max_samples=50, shuffle=True, ) classes = ["person", "dog", "cat", "bird", "car", "tree", "chair"] model = foz.load_zoo_model( "fc-clip-coco-panoptic-torch", classes=classes, ) dataset.apply_model(model, label_field="predictions") session = fo.launch_app(dataset) ```