Yolo-v7-Quantized: Optimized for Mobile Deployment
Quantized real-time object detection optimized for mobile and edge
YoloV7 is a machine learning model that predicts bounding boxes and classes of objects in an image. This model is post-training quantized to int8 using samples from the COCO dataset.
This model is an implementation of Yolo-v7-Quantized found here.
More details on model performance across various devices, can be found here.
Model Details
- Model Type: Object detection
- Model Stats:
- Model checkpoint: YoloV7 Tiny
- Input resolution: 640x640
- Number of parameters: 6.24M
- Model size: 6.23 MB
| Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model |
|---|---|---|---|---|---|---|---|---|
| Yolo-v7-Quantized | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 4.48 ms | 0 - 11 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 5.432 ms | 0 - 10 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 6.272 ms | 0 - 51 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 2.905 ms | 0 - 45 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 3.584 ms | 1 - 58 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 4.207 ms | 1 - 101 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 2.99 ms | 0 - 40 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 3.157 ms | 1 - 55 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 3.641 ms | 1 - 93 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | RB3 Gen 2 (Proxy) | QCS6490 Proxy | TFLITE | 11.977 ms | 0 - 54 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | RB3 Gen 2 (Proxy) | QCS6490 Proxy | QNN | 14.834 ms | 1 - 13 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | RB5 (Proxy) | QCS8250 Proxy | TFLITE | 56.215 ms | 15 - 54 MB | INT8 | GPU | -- |
| Yolo-v7-Quantized | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 4.477 ms | 0 - 11 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 4.431 ms | 1 - 4 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | SA7255P ADP | SA7255P | TFLITE | 19.699 ms | 0 - 32 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | SA7255P ADP | SA7255P | QNN | 20.071 ms | 1 - 10 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 4.467 ms | 0 - 11 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | SA8255 (Proxy) | SA8255P Proxy | QNN | 4.448 ms | 1 - 4 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | SA8295P ADP | SA8295P | TFLITE | 6.163 ms | 0 - 41 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | SA8295P ADP | SA8295P | QNN | 6.018 ms | 1 - 15 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 4.462 ms | 0 - 11 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | SA8650 (Proxy) | SA8650P Proxy | QNN | 4.49 ms | 1 - 4 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | SA8775P ADP | SA8775P | TFLITE | 6.197 ms | 0 - 32 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | SA8775P ADP | SA8775P | QNN | 6.464 ms | 1 - 11 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 5.152 ms | 0 - 48 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 5.085 ms | 1 - 62 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 4.895 ms | 1 - 1 MB | INT8 | NPU | -- |
| Yolo-v7-Quantized | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 8.031 ms | 8 - 8 MB | INT8 | NPU | -- |
License
- The license for the original implementation of Yolo-v7-Quantized can be found here.
- The license for the compiled assets for on-device deployment can be found here
References
- YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
- 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 API (serverless) does not yet support pytorch models for this pipeline type.