Yolo-NAS: Optimized for Mobile Deployment
Real-time object detection optimized for mobile and edge
YoloNAS is a machine learning model that predicts bounding boxes and classes of objects in an image.
This model is an implementation of Yolo-NAS found here.
More details on model performance across various devices, can be found here.
Model Details
- Model Type: Object detection
- Model Stats:
- Model checkpoint: YoloNAS Small
- Input resolution: 640x640
- Number of parameters: 12.2M
- Model size: 46.6 MB
| Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model |
|---|---|---|---|---|---|---|---|---|
| Yolo-NAS | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 7.695 ms | 0 - 18 MB | FP16 | NPU | -- |
| Yolo-NAS | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 7.909 ms | 5 - 7 MB | FP16 | NPU | -- |
| Yolo-NAS | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 8.918 ms | 1 - 64 MB | FP16 | NPU | -- |
| Yolo-NAS | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 5.197 ms | 0 - 61 MB | FP16 | NPU | -- |
| Yolo-NAS | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 5.419 ms | 5 - 24 MB | FP16 | NPU | -- |
| Yolo-NAS | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 6.025 ms | 4 - 56 MB | FP16 | NPU | -- |
| Yolo-NAS | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 5.304 ms | 0 - 47 MB | FP16 | NPU | -- |
| Yolo-NAS | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 5.403 ms | 5 - 33 MB | FP16 | NPU | -- |
| Yolo-NAS | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 5.894 ms | 4 - 40 MB | FP16 | NPU | -- |
| Yolo-NAS | SA7255P ADP | SA7255P | TFLITE | 220.428 ms | 0 - 42 MB | FP16 | NPU | -- |
| Yolo-NAS | SA7255P ADP | SA7255P | QNN | 219.093 ms | 1 - 8 MB | FP16 | NPU | -- |
| Yolo-NAS | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 7.699 ms | 0 - 21 MB | FP16 | NPU | -- |
| Yolo-NAS | SA8255 (Proxy) | SA8255P Proxy | QNN | 7.984 ms | 6 - 9 MB | FP16 | NPU | -- |
| Yolo-NAS | SA8295P ADP | SA8295P | TFLITE | 12.544 ms | 0 - 45 MB | FP16 | NPU | -- |
| Yolo-NAS | SA8295P ADP | SA8295P | QNN | 12.906 ms | 0 - 14 MB | FP16 | NPU | -- |
| Yolo-NAS | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 7.674 ms | 0 - 18 MB | FP16 | NPU | -- |
| Yolo-NAS | SA8650 (Proxy) | SA8650P Proxy | QNN | 7.993 ms | 5 - 7 MB | FP16 | NPU | -- |
| Yolo-NAS | SA8775P ADP | SA8775P | TFLITE | 13.966 ms | 0 - 44 MB | FP16 | NPU | -- |
| Yolo-NAS | SA8775P ADP | SA8775P | QNN | 14.341 ms | 0 - 10 MB | FP16 | NPU | -- |
| Yolo-NAS | QCS8275 (Proxy) | QCS8275 Proxy | TFLITE | 220.428 ms | 0 - 42 MB | FP16 | NPU | -- |
| Yolo-NAS | QCS8275 (Proxy) | QCS8275 Proxy | QNN | 219.093 ms | 1 - 8 MB | FP16 | NPU | -- |
| Yolo-NAS | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 7.698 ms | 0 - 17 MB | FP16 | NPU | -- |
| Yolo-NAS | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 7.933 ms | 5 - 7 MB | FP16 | NPU | -- |
| Yolo-NAS | QCS9075 (Proxy) | QCS9075 Proxy | TFLITE | 13.966 ms | 0 - 44 MB | FP16 | NPU | -- |
| Yolo-NAS | QCS9075 (Proxy) | QCS9075 Proxy | QNN | 14.341 ms | 0 - 10 MB | FP16 | NPU | -- |
| Yolo-NAS | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 10.711 ms | 0 - 61 MB | FP16 | NPU | -- |
| Yolo-NAS | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 12.055 ms | 5 - 39 MB | FP16 | NPU | -- |
| Yolo-NAS | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 8.518 ms | 5 - 5 MB | FP16 | NPU | -- |
| Yolo-NAS | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 9.066 ms | 21 - 21 MB | FP16 | NPU | -- |
License
- The license for the original implementation of Yolo-NAS can be found here.
- The license for the compiled assets for on-device deployment can be found here
References
- A Next-Generation, Object Detection Foundational Model generated by Deci’s Neural Architecture Search Technology
- Source Model Implementation
Community
- Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
- For questions or feedback please reach out to us.
Usage and Limitations
Model may not be used for or in connection with any of the following applications:
- Accessing essential private and public services and benefits;
- Administration of justice and democratic processes;
- Assessing or recognizing the emotional state of a person;
- Biometric and biometrics-based systems, including categorization of persons based on sensitive characteristics;
- Education and vocational training;
- Employment and workers management;
- Exploitation of the vulnerabilities of persons resulting in harmful behavior;
- General purpose social scoring;
- Law enforcement;
- Management and operation of critical infrastructure;
- Migration, asylum and border control management;
- Predictive policing;
- Real-time remote biometric identification in public spaces;
- Recommender systems of social media platforms;
- Scraping of facial images (from the internet or otherwise); and/or
- Subliminal manipulation
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.
The model cannot be deployed to the HF Inference API:
The HF Inference API does not support object-detection models for pytorch library.