| --- |
| library_name: pytorch |
| license: other |
| tags: |
| - android |
| pipeline_tag: object-detection |
|
|
| --- |
| |
|  |
|
|
| # RF-DETR: Optimized for Qualcomm Devices |
|
|
| DETR is a machine learning model that can detect objects (trained on COCO dataset). |
|
|
| This is based on the implementation of RF-DETR found [here](https://github.com/roboflow/rf-detr). |
| This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/rf_detr) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary). |
|
|
| Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device. |
|
|
| ## Getting Started |
| There are two ways to deploy this model on your device: |
|
|
| ### Option 1: Download Pre-Exported Models |
|
|
| Below are pre-exported model assets ready for deployment. |
|
|
| | Runtime | Precision | Chipset | SDK Versions | Download | |
| |---|---|---|---|---| |
| | ONNX | float | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/rf_detr/releases/v0.60.0/rf_detr-onnx-float.zip) |
| | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/rf_detr/releases/v0.60.0/rf_detr-qnn_dlc-float.zip) |
| |
| For more device-specific assets and performance metrics, visit **[RF-DETR on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/rf_detr)**. |
| |
| |
| ### Option 2: Export with Custom Configurations |
| |
| Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/rf_detr) Python library to compile and export the model with your own: |
| - Custom weights (e.g., fine-tuned checkpoints) |
| - Custom input shapes |
| - Target device and runtime configurations |
| |
| This option is ideal if you need to customize the model beyond the default configuration provided here. |
| |
| See our repository for [RF-DETR on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/rf_detr) for usage instructions. |
| |
| ## Model Details |
| |
| **Model Type:** Model_use_case.object_detection |
|
|
| **Model Stats:** |
| - Input resolution: 512x512 |
| - Model checkpoint: RF-DETR-small |
| - Model size (float): 109 MB |
| - Number of parameters: 28.5M |
| - Supported variants: nano (384x384), small (512x512), medium (576x576), base (560x560) |
|
|
| ## Performance Summary |
| | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
| |---|---|---|---|---|---|--- |
| | RF-DETR | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 30.758 ms | 0 - 408 MB | NPU |
| | RF-DETR | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 99.23 ms | 13 - 431 MB | NPU |
| | RF-DETR | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 51.536 ms | 9 - 16 MB | NPU |
| | RF-DETR | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 41.839 ms | 6 - 10 MB | NPU |
| | RF-DETR | ONNX | float | Qualcomm® QCS8450 | 99.23 ms | 13 - 431 MB | NPU |
| | RF-DETR | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 50.38 ms | 11 - 17 MB | NPU |
| | RF-DETR | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 24.011 ms | 10 - 332 MB | NPU |
| | RF-DETR | ONNX | float | Snapdragon® 8 Elite Mobile | 24.011 ms | 10 - 332 MB | NPU |
| | RF-DETR | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 20.115 ms | 11 - 364 MB | NPU |
| | RF-DETR | QNN_DLC | float | Snapdragon® X2 Elite | 25.273 ms | 3 - 3 MB | NPU |
| | RF-DETR | QNN_DLC | float | Snapdragon® X Elite | 50.732 ms | 3 - 3 MB | NPU |
| | RF-DETR | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 36.375 ms | 0 - 472 MB | NPU |
| | RF-DETR | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 99.537 ms | 3 - 477 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 56.325 ms | 3 - 8 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 135.464 ms | 1 - 358 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 49.887 ms | 3 - 6 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® SA8775P | 56.777 ms | 1 - 385 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® SA8650P | 56.777 ms | 1 - 385 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® SA8255P | 56.777 ms | 1 - 385 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® QCS8450 | 99.537 ms | 3 - 477 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 57.242 ms | 5 - 10 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 50.732 ms | 3 - 3 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 28.162 ms | 0 - 407 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® SA7255P | 135.464 ms | 1 - 358 MB | NPU |
| | RF-DETR | QNN_DLC | float | Qualcomm® SA8295P | 81.197 ms | 0 - 365 MB | NPU |
| | RF-DETR | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 28.162 ms | 0 - 407 MB | NPU |
| | RF-DETR | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 23.364 ms | 3 - 437 MB | NPU |
|
|
| ## License |
| * The license for the original implementation of RF-DETR can be found |
| [here](https://github.com/roboflow/rf-detr/blob/develop/LICENSE). |
|
|
| ## References |
| * [RF-DETR A SOTA Real-Time Object Detection Model](https://blog.roboflow.com/rf-detr/) |
| * [Source Model Implementation](https://github.com/roboflow/rf-detr) |
|
|
| ## Community |
| * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI. |
| * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com). |
|
|
