Yolo-v4: Optimized for Qualcomm Devices

YoloV4 is a machine learning model that predicts bounding boxes and classes of objects in an image.

This is based on the implementation of Yolo-v4 found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up 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 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-v4 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.object_detection

Model Stats:

• Input resolution: 416x416
• Model checkpoint: YoloV4 Tiny
• Model size (float): 23.1 MB
• Model size (w8a16): 6.07 MB
• Number of parameters: 6.06M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
Yolo-v4	ONNX	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	0.833 ms	0 - 29 MB	NPU
Yolo-v4	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	0.975 ms	0 - 29 MB	NPU
Yolo-v4	ONNX	float	Snapdragon® X2 Elite	0.964 ms	2 - 2 MB	NPU
Yolo-v4	ONNX	float	Snapdragon® X Elite	1.896 ms	10 - 10 MB	NPU
Yolo-v4	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.27 ms	0 - 43 MB	NPU
Yolo-v4	ONNX	float	Snapdragon® 8 Gen 1 Mobile	2.894 ms	1 - 51 MB	NPU
Yolo-v4	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	2.891 ms	2 - 6 MB	NPU
Yolo-v4	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.843 ms	0 - 12 MB	NPU
Yolo-v4	ONNX	float	Qualcomm® QCS8450	2.894 ms	1 - 51 MB	NPU
Yolo-v4	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	2.744 ms	2 - 5 MB	NPU
Yolo-v4	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	1.896 ms	10 - 10 MB	NPU
Yolo-v4	ONNX	float	Qualcomm® Dragonwing™ Q-8750	0.975 ms	0 - 29 MB	NPU
Yolo-v4	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	0.615 ms	0 - 35 MB	NPU
Yolo-v4	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	0.769 ms	0 - 35 MB	NPU
Yolo-v4	ONNX	w8a16	Snapdragon® X2 Elite	0.703 ms	2 - 2 MB	NPU
Yolo-v4	ONNX	w8a16	Snapdragon® X Elite	1.501 ms	5 - 5 MB	NPU
Yolo-v4	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	0.987 ms	0 - 56 MB	NPU
Yolo-v4	ONNX	w8a16	Snapdragon® 8 Gen 1 Mobile	2.079 ms	0 - 61 MB	NPU
Yolo-v4	ONNX	w8a16	Qualcomm® Dragonwing™ QCS6490	14.81 ms	16 - 19 MB	NPU
Yolo-v4	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-8275	1.493 ms	0 - 4 MB	NPU
Yolo-v4	ONNX	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.454 ms	0 - 61 MB	NPU
Yolo-v4	ONNX	w8a16	Qualcomm® QCS8450	2.079 ms	0 - 61 MB	NPU
Yolo-v4	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-9075	1.571 ms	0 - 3 MB	NPU
Yolo-v4	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-X7181	1.501 ms	5 - 5 MB	NPU
Yolo-v4	ONNX	w8a16	Qualcomm® Dragonwing™ Q-6690	17.434 ms	16 - 167 MB	NPU
Yolo-v4	ONNX	w8a16	Qualcomm® Dragonwing™ Q-7790	4.439 ms	17 - 160 MB	NPU
Yolo-v4	ONNX	w8a16	Qualcomm® Dragonwing™ Q-8750	0.769 ms	0 - 35 MB	NPU
Yolo-v4	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	4.439 ms	17 - 160 MB	NPU
Yolo-v4	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	0.696 ms	2 - 31 MB	NPU
Yolo-v4	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	0.865 ms	0 - 31 MB	NPU
Yolo-v4	QNN_DLC	float	Snapdragon® X2 Elite	0.951 ms	2 - 2 MB	NPU
Yolo-v4	QNN_DLC	float	Snapdragon® X Elite	1.719 ms	2 - 2 MB	NPU
Yolo-v4	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.121 ms	0 - 39 MB	NPU
Yolo-v4	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	2.297 ms	2 - 48 MB	NPU
Yolo-v4	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	2.603 ms	2 - 7 MB	NPU
Yolo-v4	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.544 ms	2 - 4 MB	NPU
Yolo-v4	QNN_DLC	float	Qualcomm® SA8775P	2.505 ms	0 - 30 MB	NPU
Yolo-v4	QNN_DLC	float	Qualcomm® SA8650P	2.505 ms	0 - 30 MB	NPU
Yolo-v4	QNN_DLC	float	Qualcomm® SA8255P	2.505 ms	0 - 30 MB	NPU
Yolo-v4	QNN_DLC	float	Qualcomm® QCS8450	2.297 ms	2 - 48 MB	NPU
Yolo-v4	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	2.367 ms	4 - 8 MB	NPU
Yolo-v4	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	1.719 ms	2 - 2 MB	NPU
Yolo-v4	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	0.865 ms	0 - 31 MB	NPU
Yolo-v4	QNN_DLC	float	Qualcomm® SA8295P	2.641 ms	0 - 27 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	0.476 ms	1 - 32 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	0.632 ms	0 - 36 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Snapdragon® X2 Elite	0.729 ms	1 - 1 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Snapdragon® X Elite	1.261 ms	1 - 1 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	0.826 ms	0 - 51 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Snapdragon® 8 Gen 1 Mobile	1.527 ms	0 - 57 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® Dragonwing™ QCS6490	3.835 ms	1 - 4 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-8275	1.224 ms	1 - 5 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.078 ms	1 - 2 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® SA8775P	1.279 ms	1 - 36 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® SA8650P	1.279 ms	1 - 36 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® SA8255P	1.279 ms	1 - 36 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® QCS8450	1.527 ms	0 - 57 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-9075	1.142 ms	1 - 4 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-X7181	1.261 ms	1 - 1 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-6690	11.789 ms	1 - 143 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-7790	1.422 ms	0 - 37 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-8750	0.632 ms	0 - 36 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Qualcomm® SA8295P	1.758 ms	0 - 31 MB	NPU
Yolo-v4	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	1.422 ms	0 - 37 MB	NPU
Yolo-v4	TFLITE	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	0.689 ms	0 - 30 MB	NPU
Yolo-v4	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	0.872 ms	0 - 34 MB	NPU
Yolo-v4	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1.135 ms	0 - 51 MB	NPU
Yolo-v4	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	2.31 ms	0 - 60 MB	NPU
Yolo-v4	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	2.626 ms	0 - 17 MB	NPU
Yolo-v4	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.51 ms	0 - 2 MB	NPU
Yolo-v4	TFLITE	float	Qualcomm® SA8775P	2.547 ms	0 - 33 MB	NPU
Yolo-v4	TFLITE	float	Qualcomm® SA8650P	2.547 ms	0 - 33 MB	NPU
Yolo-v4	TFLITE	float	Qualcomm® SA8255P	2.547 ms	0 - 33 MB	NPU
Yolo-v4	TFLITE	float	Qualcomm® QCS8450	2.31 ms	0 - 60 MB	NPU
Yolo-v4	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	2.378 ms	0 - 16 MB	NPU
Yolo-v4	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	0.872 ms	0 - 34 MB	NPU
Yolo-v4	TFLITE	float	Qualcomm® SA7255P	7.908 ms	0 - 31 MB	NPU
Yolo-v4	TFLITE	float	Qualcomm® SA8295P	2.663 ms	0 - 28 MB	NPU

License

• The license for the original implementation of Yolo-v4 can be found here.

References

• YOLOv4: Optimal Speed and Accuracy of Object Detection
• 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.