BGNet: Optimized for Qualcomm Devices

BGNet or Boundary-Guided Network, is a model designed for camouflaged object detection. It leverages edge semantics to enhance the representation learning process, making it more effective at identifying objects that blend into their surroundings

This is based on the implementation of BGNet 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 BGNet on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.semantic_segmentation

Model Stats:

• Model checkpoint: BGNet
• Input resolution: 416x416
• Number of parameters: 77.8M
• Model size (float): 297 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
BGNet	ONNX	float	Snapdragon® X Elite	19.578 ms	153 - 153 MB	NPU
BGNet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	13.995 ms	2 - 319 MB	NPU
BGNet	ONNX	float	Qualcomm® QCS8550 (Proxy)	19.168 ms	0 - 522 MB	NPU
BGNet	ONNX	float	Qualcomm® QCS9075	36.326 ms	3 - 7 MB	NPU
BGNet	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	11.407 ms	3 - 245 MB	NPU
BGNet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	9.095 ms	3 - 256 MB	NPU
BGNet	ONNX	float	Snapdragon® X2 Elite	9.934 ms	154 - 154 MB	NPU
BGNet	ONNX	w8a16	Snapdragon® X Elite	12.145 ms	78 - 78 MB	NPU
BGNet	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	8.575 ms	0 - 405 MB	NPU
BGNet	ONNX	w8a16	Qualcomm® QCS6490	2704.023 ms	340 - 388 MB	CPU
BGNet	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	11.612 ms	0 - 85 MB	NPU
BGNet	ONNX	w8a16	Qualcomm® QCS9075	13.88 ms	1 - 4 MB	NPU
BGNet	ONNX	w8a16	Qualcomm® QCM6690	1290.238 ms	289 - 304 MB	CPU
BGNet	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	6.946 ms	2 - 260 MB	NPU
BGNet	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	1288.239 ms	247 - 258 MB	CPU
BGNet	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	4.894 ms	2 - 271 MB	NPU
BGNet	ONNX	w8a16	Snapdragon® X2 Elite	6.22 ms	78 - 78 MB	NPU
BGNet	ONNX	w8a8	Snapdragon® X Elite	6.965 ms	77 - 77 MB	NPU
BGNet	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	4.55 ms	0 - 338 MB	NPU
BGNet	ONNX	w8a8	Qualcomm® QCS6490	437.421 ms	48 - 151 MB	CPU
BGNet	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	6.779 ms	0 - 83 MB	NPU
BGNet	ONNX	w8a8	Qualcomm® QCS9075	7.246 ms	0 - 4 MB	NPU
BGNet	ONNX	w8a8	Qualcomm® QCM6690	397.068 ms	28 - 43 MB	CPU
BGNet	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	3.87 ms	0 - 228 MB	NPU
BGNet	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	337.047 ms	0 - 14 MB	CPU
BGNet	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	3.231 ms	0 - 230 MB	NPU
BGNet	ONNX	w8a8	Snapdragon® X2 Elite	2.993 ms	77 - 77 MB	NPU
BGNet	QNN_DLC	float	Snapdragon® X Elite	19.985 ms	2 - 2 MB	NPU
BGNet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	14.312 ms	2 - 314 MB	NPU
BGNet	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	115.242 ms	2 - 238 MB	NPU
BGNet	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	19.475 ms	2 - 4 MB	NPU
BGNet	QNN_DLC	float	Qualcomm® SA8775P	145.877 ms	2 - 237 MB	NPU
BGNet	QNN_DLC	float	Qualcomm® QCS9075	37.186 ms	4 - 8 MB	NPU
BGNet	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	39.181 ms	0 - 258 MB	NPU
BGNet	QNN_DLC	float	Qualcomm® SA7255P	115.242 ms	2 - 238 MB	NPU
BGNet	QNN_DLC	float	Qualcomm® SA8295P	33.952 ms	2 - 195 MB	NPU
BGNet	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	11.476 ms	2 - 237 MB	NPU
BGNet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	8.77 ms	2 - 245 MB	NPU
BGNet	QNN_DLC	float	Snapdragon® X2 Elite	10.067 ms	2 - 2 MB	NPU
BGNet	QNN_DLC	w8a16	Snapdragon® X Elite	13.023 ms	1 - 1 MB	NPU
BGNet	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	9.092 ms	1 - 392 MB	NPU
BGNet	QNN_DLC	w8a16	Qualcomm® QCS6490	61.652 ms	1 - 4 MB	NPU
BGNet	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	36.561 ms	1 - 258 MB	NPU
BGNet	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	12.476 ms	1 - 3 MB	NPU
BGNet	QNN_DLC	w8a16	Qualcomm® SA8775P	12.814 ms	1 - 259 MB	NPU
BGNet	QNN_DLC	w8a16	Qualcomm® QCS9075	15.092 ms	1 - 4 MB	NPU
BGNet	QNN_DLC	w8a16	Qualcomm® QCM6690	276.575 ms	1 - 363 MB	NPU
BGNet	QNN_DLC	w8a16	Qualcomm® QCS8450 (Proxy)	20.864 ms	0 - 376 MB	NPU
BGNet	QNN_DLC	w8a16	Qualcomm® SA7255P	36.561 ms	1 - 258 MB	NPU
BGNet	QNN_DLC	w8a16	Qualcomm® SA8295P	20.527 ms	1 - 258 MB	NPU
BGNet	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	6.909 ms	1 - 252 MB	NPU
BGNet	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	23.08 ms	1 - 344 MB	NPU
BGNet	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	5.907 ms	1 - 258 MB	NPU
BGNet	QNN_DLC	w8a16	Snapdragon® X2 Elite	7.068 ms	1 - 1 MB	NPU
BGNet	QNN_DLC	w8a8	Snapdragon® X Elite	6.419 ms	0 - 0 MB	NPU
BGNet	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	4.373 ms	0 - 319 MB	NPU
BGNet	QNN_DLC	w8a8	Qualcomm® QCS6490	26.899 ms	2 - 5 MB	NPU
BGNet	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	17.839 ms	1 - 213 MB	NPU
BGNet	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	6.183 ms	1 - 20 MB	NPU
BGNet	QNN_DLC	w8a8	Qualcomm® SA8775P	6.56 ms	1 - 213 MB	NPU
BGNet	QNN_DLC	w8a8	Qualcomm® QCS9075	7.368 ms	1 - 3 MB	NPU
BGNet	QNN_DLC	w8a8	Qualcomm® QCM6690	100.396 ms	1 - 293 MB	NPU
BGNet	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	9.573 ms	0 - 321 MB	NPU
BGNet	QNN_DLC	w8a8	Qualcomm® SA7255P	17.839 ms	1 - 213 MB	NPU
BGNet	QNN_DLC	w8a8	Qualcomm® SA8295P	10.016 ms	0 - 215 MB	NPU
BGNet	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	3.529 ms	1 - 212 MB	NPU
BGNet	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	10.467 ms	1 - 287 MB	NPU
BGNet	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	2.855 ms	1 - 213 MB	NPU
BGNet	QNN_DLC	w8a8	Snapdragon® X2 Elite	3.15 ms	1 - 1 MB	NPU
BGNet	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	14.464 ms	1 - 455 MB	NPU
BGNet	TFLITE	float	Qualcomm® QCS8275 (Proxy)	113.916 ms	1 - 299 MB	NPU
BGNet	TFLITE	float	Qualcomm® QCS8550 (Proxy)	19.864 ms	1 - 3 MB	NPU
BGNet	TFLITE	float	Qualcomm® SA8775P	31.991 ms	1 - 299 MB	NPU
BGNet	TFLITE	float	Qualcomm® QCS9075	34.931 ms	1 - 159 MB	NPU
BGNet	TFLITE	float	Qualcomm® QCS8450 (Proxy)	37.053 ms	1 - 416 MB	NPU
BGNet	TFLITE	float	Qualcomm® SA7255P	113.916 ms	1 - 299 MB	NPU
BGNet	TFLITE	float	Qualcomm® SA8295P	32.769 ms	1 - 261 MB	NPU
BGNet	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	11.576 ms	0 - 299 MB	NPU
BGNet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	8.65 ms	1 - 306 MB	NPU

License

• The license for the original implementation of BGNet can be found here.

References

• BGNet: Boundary-Guided Camouflaged Object Detection (IJCAI 2022)
• 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.