| license: apache-2.0 | |
| tags: | |
| - coreml | |
| - object-detection | |
| - real-time | |
| - rfdetr | |
| - computer-vision | |
| # Rfdetr Nano Fp16 Coreml | |
| RF-DETR Nano model exported to CoreML format (FP16) for Apple devices. Fastest inference with 384×384 resolution. | |
| ## Model Details | |
| - **Model Type**: Object Detection | |
| - **Framework**: CoreML | |
| - **Precision**: FP16 | |
| - **Input Resolution**: 384×384 | |
| - **Performance**: ~56 FPS on iPhone 13 Mini | |
| - **Inference Time**: 17.96 ms | |
| ## Usage | |
| This model can be used for real-time object detection on Apple devices (iOS, macOS, watchOS, tvOS). | |
| ## Custom conversion to CoreML | |
| Check the repo https://github.com/pzoltowski/rf-detr-coreml | |
| ## Performance | |
| | Model | Resolution | Inference Time (ms) | FPS | | |
| |-------|------------|---------------------|-----| | |
| | RF-DETR-Nano | 384×384 | 17.96 | ~56 | | |
| | RF-DETR-Small | 512×512 | 27.07 | ~37 | | |
| | RF-DETR-Medium | 576×576 | 41.85 | ~24 | | |
| *Performance measured on iPhone 13 Mini using Xcode performance tools with FP16 precision.* | |
| ``` | |