README.md

---
library_name: pytorch
license: other
tags:
- bu_auto
- real_time
- android
pipeline_tag: object-detection

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yolov3/web-assets/model_demo.png)

# Yolo-v3: Optimized for Qualcomm Devices

YoloV3 is a machine learning model that predicts bounding boxes and classes of objects in an image.

This is based on the implementation of Yolo-v3 found [here](https://github.com/ultralytics/yolov3/tree/v8).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/yolov3) 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
Due to licensing restrictions, we cannot distribute pre-exported model assets for this model.
Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/yolov3) 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

See our repository for [Yolo-v3 on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/yolov3) for usage instructions.


## Model Details

**Model Type:** Model_use_case.object_detection

**Model Stats:**
- Model checkpoint: YoloV3 Tiny
- Input resolution: 416p (416x416)
- Number of parameters: 11.5M
- Model size (float): 43.9 MB
- Model size (w8a16): 16.9 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| Yolo-v3 | ONNX | float | Snapdragon® X Elite | 4.281 ms | 19 - 19 MB | NPU
| Yolo-v3 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 3.31 ms | 50 - 167 MB | NPU
| Yolo-v3 | ONNX | float | Qualcomm® QCS8550 (Proxy) | 4.258 ms | 0 - 171 MB | NPU
| Yolo-v3 | ONNX | float | Qualcomm® QCS9075 | 6.603 ms | 5 - 7 MB | NPU
| Yolo-v3 | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.805 ms | 1 - 99 MB | NPU
| Yolo-v3 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.02 ms | 0 - 101 MB | NPU
| Yolo-v3 | ONNX | w8a16 | Snapdragon® X Elite | 3.987 ms | 10 - 10 MB | NPU
| Yolo-v3 | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 2.679 ms | 0 - 151 MB | NPU
| Yolo-v3 | ONNX | w8a16 | Qualcomm® QCS6490 | 528.456 ms | 210 - 216 MB | CPU
| Yolo-v3 | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 3.964 ms | 0 - 18 MB | NPU
| Yolo-v3 | ONNX | w8a16 | Qualcomm® QCS9075 | 4.235 ms | 2 - 5 MB | NPU
| Yolo-v3 | ONNX | w8a16 | Qualcomm® QCM6690 | 246.489 ms | 218 - 225 MB | CPU
| Yolo-v3 | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 2.074 ms | 0 - 119 MB | NPU
| Yolo-v3 | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 214.463 ms | 215 - 222 MB | CPU
| Yolo-v3 | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 1.567 ms | 0 - 119 MB | NPU
| Yolo-v3 | QNN_DLC | float | Snapdragon® X Elite | 3.695 ms | 5 - 5 MB | NPU
| Yolo-v3 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 2.785 ms | 0 - 159 MB | NPU
| Yolo-v3 | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 20.116 ms | 1 - 146 MB | NPU
| Yolo-v3 | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 3.385 ms | 5 - 111 MB | NPU
| Yolo-v3 | QNN_DLC | float | Qualcomm® QCS9075 | 5.756 ms | 7 - 13 MB | NPU
| Yolo-v3 | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 10.075 ms | 4 - 173 MB | NPU
| Yolo-v3 | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.122 ms | 0 - 146 MB | NPU
| Yolo-v3 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.653 ms | 5 - 153 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Snapdragon® X Elite | 4.028 ms | 2 - 2 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 2.557 ms | 2 - 77 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Qualcomm® QCS6490 | 11.727 ms | 4 - 8 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 9.181 ms | 1 - 46 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 3.673 ms | 2 - 112 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 3.926 ms | 3 - 7 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 35.002 ms | 2 - 174 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 4.355 ms | 2 - 78 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 1.897 ms | 0 - 46 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 4.73 ms | 2 - 175 MB | NPU
| Yolo-v3 | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 1.366 ms | 2 - 53 MB | NPU
| Yolo-v3 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 2.719 ms | 0 - 189 MB | NPU
| Yolo-v3 | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 19.991 ms | 0 - 149 MB | NPU
| Yolo-v3 | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 3.21 ms | 0 - 2 MB | NPU
| Yolo-v3 | TFLITE | float | Qualcomm® QCS9075 | 5.637 ms | 0 - 30 MB | NPU
| Yolo-v3 | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 9.944 ms | 0 - 194 MB | NPU
| Yolo-v3 | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.102 ms | 0 - 144 MB | NPU
| Yolo-v3 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.591 ms | 0 - 151 MB | NPU

## License
* The license for the original implementation of Yolo-v3 can be found
  [here](https://github.com/ultralytics/yolov3/blob/v8/LICENSE).

## References
* [YOLOv3: An Incremental Improvement](https://arxiv.org/abs/1804.02767)
* [Source Model Implementation](https://github.com/ultralytics/yolov3/tree/v8)

## 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).