DDRNet23-Slim: Optimized for Qualcomm Devices

DDRNet23Slim is a machine learning model that segments an image into semantic classes, specifically designed for road-based scenes. It is designed for the application of self-driving cars.

This is based on the implementation of DDRNet23-Slim 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

There are two ways to deploy this model on your device:

Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

Runtime	Precision	Chipset	SDK Versions	Download
ONNX	float	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
ONNX	w8a8	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	Download
TFLITE	w8a8	Universal	QAIRT 2.45	Download

For more device-specific assets and performance metrics, visit DDRNet23-Slim on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

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

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for DDRNet23-Slim on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.semantic_segmentation

Model Stats:

Model checkpoint: DDRNet23s_imagenet.pth
Inference latency: RealTime
Input resolution: 2048x1024
Number of output classes: 19
Number of parameters: 6.13M
Model size (float): 21.7 MB
Model size (w8a8): 6.11 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
DDRNet23-Slim	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	10.47 ms	31 - 256 MB	NPU
DDRNet23-Slim	ONNX	float	Snapdragon® X2 Elite	10.888 ms	22 - 22 MB	NPU
DDRNet23-Slim	ONNX	float	Snapdragon® X Elite	28.119 ms	24 - 24 MB	NPU
DDRNet23-Slim	ONNX	float	Snapdragon® 8 Gen 3 Mobile	19.79 ms	32 - 308 MB	NPU
DDRNet23-Slim	ONNX	float	Qualcomm® QCS8550 (Proxy)	28.67 ms	24 - 28 MB	NPU
DDRNet23-Slim	ONNX	float	Qualcomm® QCS9075	39.311 ms	24 - 51 MB	NPU
DDRNet23-Slim	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	13.629 ms	5 - 203 MB	NPU
DDRNet23-Slim	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	44.509 ms	56 - 247 MB	NPU
DDRNet23-Slim	ONNX	w8a8	Snapdragon® X2 Elite	43.943 ms	109 - 109 MB	NPU
DDRNet23-Slim	ONNX	w8a8	Snapdragon® X Elite	87.005 ms	109 - 109 MB	NPU
DDRNet23-Slim	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	43.323 ms	92 - 339 MB	NPU
DDRNet23-Slim	ONNX	w8a8	Qualcomm® QCS6490	299.285 ms	199 - 216 MB	CPU
DDRNet23-Slim	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	57.833 ms	81 - 92 MB	NPU
DDRNet23-Slim	ONNX	w8a8	Qualcomm® QCS9075	63.814 ms	87 - 89 MB	NPU
DDRNet23-Slim	ONNX	w8a8	Qualcomm® QCM6690	266.633 ms	201 - 209 MB	CPU
DDRNet23-Slim	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	42.032 ms	83 - 270 MB	NPU
DDRNet23-Slim	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	250.631 ms	139 - 148 MB	CPU
DDRNet23-Slim	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	10.391 ms	12 - 249 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Snapdragon® X2 Elite	11.73 ms	24 - 24 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Snapdragon® X Elite	33.791 ms	24 - 24 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	22.104 ms	24 - 298 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	98.382 ms	24 - 220 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	32.726 ms	24 - 26 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Qualcomm® SA8775P	40.32 ms	24 - 221 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Qualcomm® QCS9075	53.286 ms	24 - 52 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	66.995 ms	6 - 280 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Qualcomm® SA7255P	98.382 ms	24 - 220 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Qualcomm® SA8295P	43.449 ms	24 - 228 MB	NPU
DDRNet23-Slim	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	15.617 ms	16 - 237 MB	NPU
DDRNet23-Slim	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	10.363 ms	2 - 241 MB	NPU
DDRNet23-Slim	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	22.518 ms	2 - 285 MB	NPU
DDRNet23-Slim	TFLITE	float	Qualcomm® QCS8275 (Proxy)	98.289 ms	0 - 202 MB	NPU
DDRNet23-Slim	TFLITE	float	Qualcomm® QCS8550 (Proxy)	33.206 ms	2 - 4 MB	NPU
DDRNet23-Slim	TFLITE	float	Qualcomm® SA8775P	40.402 ms	2 - 205 MB	NPU
DDRNet23-Slim	TFLITE	float	Qualcomm® QCS9075	53.887 ms	0 - 41 MB	NPU
DDRNet23-Slim	TFLITE	float	Qualcomm® QCS8450 (Proxy)	67.023 ms	3 - 287 MB	NPU
DDRNet23-Slim	TFLITE	float	Qualcomm® SA7255P	98.289 ms	0 - 202 MB	NPU
DDRNet23-Slim	TFLITE	float	Qualcomm® SA8295P	43.484 ms	2 - 215 MB	NPU
DDRNet23-Slim	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	15.381 ms	2 - 225 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	21.521 ms	1 - 231 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	37.132 ms	1 - 248 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Qualcomm® QCS6490	193.066 ms	10 - 78 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	95.072 ms	1 - 200 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	48.93 ms	1 - 3 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Qualcomm® SA8775P	49.656 ms	1 - 201 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Qualcomm® QCS9075	51.53 ms	0 - 15 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Qualcomm® QCM6690	212.826 ms	10 - 228 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	56.003 ms	0 - 245 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Qualcomm® SA7255P	95.072 ms	1 - 200 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Qualcomm® SA8295P	56.125 ms	1 - 204 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	68.889 ms	1 - 221 MB	NPU
DDRNet23-Slim	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	63.166 ms	11 - 211 MB	NPU

License

The license for the original implementation of DDRNet23-Slim can be found here.

References

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.

Downloads last month: -; Downloads are not tracked for this model. How to track

Paper for qualcomm/DDRNet23-Slim

Deep Dual-resolution Networks for Real-time and Accurate Semantic Segmentation of Road Scenes

Paper • 2101.06085 • Published Jan 15, 2021