New models in the FiftyOne model zoo: Broader coverage, same simple workflow

FiftyOne has expanded its Model Zoo with a variety of new model families and plugins, enhancing its capabilities for image classification, object detection, semantic segmentation, and medical zero-shot tasks. These updates integrate with the Hugging Face Transformers, allowing users to load popular model checkpoints, apply them to datasets, and analyze results using FiftyOne's interactive App. This integration maintains a consistent workflow for evaluation and comparison, offering tools like confusion matrices and PR curves to facilitate detailed analysis. The new models include ConvNeXt, EfficientNet, Swin V2 for classification, RT-DETR v2, D-FINE for detection, SegFormer for segmentation, and several medical zero-shot vision-language models, all of which can be run using efficient device handling for both GPU and CPU. Compared to other platforms like Hugging Face and Ultralytics YOLO, FiftyOne provides a dataset-centric, model-agnostic environment that emphasizes data curation and model evaluation, making it easier to compare models and refine data-driven decisions.