Midas-V2: Optimized for Qualcomm Devices

Midas is designed for estimating depth at each point in an image.

This is based on the implementation of Midas-V2 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 2.43	Download
QNN_DLC	w8a8	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.19.1	Download
TFLITE	w8a8	Universal	QAIRT 2.43, TFLite 2.19.1	Download

For more device-specific assets and performance metrics, visit Midas-V2 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 Midas-V2 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.depth_estimation

Model Stats:

• Model checkpoint: MiDaS_small
• Input resolution: 256x256
• Number of parameters: 16.6M
• Model size (float): 63.2 MB
• Model size (w8a8): 16.9 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
Midas-V2	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1.292 ms	1 - 162 MB	NPU
Midas-V2	ONNX	float	Snapdragon® X2 Elite	1.318 ms	35 - 35 MB	NPU
Midas-V2	ONNX	float	Snapdragon® X Elite	2.943 ms	35 - 35 MB	NPU
Midas-V2	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.924 ms	0 - 201 MB	NPU
Midas-V2	ONNX	float	Qualcomm® QCS8550 (Proxy)	2.732 ms	0 - 42 MB	NPU
Midas-V2	ONNX	float	Qualcomm® QCS9075	3.878 ms	1 - 4 MB	NPU
Midas-V2	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	1.477 ms	0 - 164 MB	NPU
Midas-V2	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.654 ms	0 - 58 MB	NPU
Midas-V2	ONNX	w8a8	Snapdragon® X2 Elite	0.632 ms	20 - 20 MB	NPU
Midas-V2	ONNX	w8a8	Snapdragon® X Elite	1.517 ms	18 - 18 MB	NPU
Midas-V2	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	0.945 ms	0 - 95 MB	NPU
Midas-V2	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	1.33 ms	0 - 4 MB	NPU
Midas-V2	ONNX	w8a8	Qualcomm® QCS9075	1.514 ms	0 - 3 MB	NPU
Midas-V2	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.758 ms	0 - 53 MB	NPU
Midas-V2	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	1.27 ms	1 - 166 MB	NPU
Midas-V2	QNN_DLC	float	Snapdragon® X2 Elite	1.591 ms	1 - 1 MB	NPU
Midas-V2	QNN_DLC	float	Snapdragon® X Elite	3.297 ms	1 - 1 MB	NPU
Midas-V2	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	2.134 ms	0 - 196 MB	NPU
Midas-V2	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	12.111 ms	1 - 159 MB	NPU
Midas-V2	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	3.05 ms	1 - 3 MB	NPU
Midas-V2	QNN_DLC	float	Qualcomm® SA8775P	4.279 ms	1 - 160 MB	NPU
Midas-V2	QNN_DLC	float	Qualcomm® QCS9075	4.496 ms	1 - 3 MB	NPU
Midas-V2	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	6.853 ms	0 - 193 MB	NPU
Midas-V2	QNN_DLC	float	Qualcomm® SA7255P	12.111 ms	1 - 159 MB	NPU
Midas-V2	QNN_DLC	float	Qualcomm® SA8295P	5.491 ms	1 - 161 MB	NPU
Midas-V2	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	1.614 ms	1 - 161 MB	NPU
Midas-V2	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.566 ms	0 - 50 MB	NPU
Midas-V2	QNN_DLC	w8a8	Snapdragon® X2 Elite	0.723 ms	0 - 0 MB	NPU
Midas-V2	QNN_DLC	w8a8	Snapdragon® X Elite	1.479 ms	0 - 0 MB	NPU
Midas-V2	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	0.923 ms	0 - 81 MB	NPU
Midas-V2	QNN_DLC	w8a8	Qualcomm® QCS6490	4.381 ms	0 - 2 MB	NPU
Midas-V2	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	2.87 ms	0 - 46 MB	NPU
Midas-V2	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	1.336 ms	0 - 12 MB	NPU
Midas-V2	QNN_DLC	w8a8	Qualcomm® SA8775P	1.591 ms	0 - 49 MB	NPU
Midas-V2	QNN_DLC	w8a8	Qualcomm® QCS9075	1.46 ms	0 - 2 MB	NPU
Midas-V2	QNN_DLC	w8a8	Qualcomm® QCM6690	8.916 ms	0 - 173 MB	NPU
Midas-V2	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	1.847 ms	0 - 80 MB	NPU
Midas-V2	QNN_DLC	w8a8	Qualcomm® SA7255P	2.87 ms	0 - 46 MB	NPU
Midas-V2	QNN_DLC	w8a8	Qualcomm® SA8295P	2.257 ms	0 - 45 MB	NPU
Midas-V2	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.684 ms	0 - 51 MB	NPU
Midas-V2	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	1.598 ms	0 - 170 MB	NPU
Midas-V2	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1.273 ms	0 - 184 MB	NPU
Midas-V2	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	2.126 ms	0 - 233 MB	NPU
Midas-V2	TFLITE	float	Qualcomm® QCS8275 (Proxy)	12.202 ms	0 - 179 MB	NPU
Midas-V2	TFLITE	float	Qualcomm® QCS8550 (Proxy)	3.034 ms	0 - 3 MB	NPU
Midas-V2	TFLITE	float	Qualcomm® SA8775P	4.293 ms	0 - 180 MB	NPU
Midas-V2	TFLITE	float	Qualcomm® QCS9075	4.448 ms	0 - 39 MB	NPU
Midas-V2	TFLITE	float	Qualcomm® QCS8450 (Proxy)	6.884 ms	0 - 217 MB	NPU
Midas-V2	TFLITE	float	Qualcomm® SA7255P	12.202 ms	0 - 179 MB	NPU
Midas-V2	TFLITE	float	Qualcomm® SA8295P	5.496 ms	0 - 158 MB	NPU
Midas-V2	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	1.618 ms	0 - 177 MB	NPU
Midas-V2	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.489 ms	0 - 48 MB	NPU
Midas-V2	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	0.759 ms	0 - 85 MB	NPU
Midas-V2	TFLITE	w8a8	Qualcomm® QCS6490	3.649 ms	0 - 28 MB	NPU
Midas-V2	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	2.486 ms	0 - 44 MB	NPU
Midas-V2	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	1.067 ms	0 - 3 MB	NPU
Midas-V2	TFLITE	w8a8	Qualcomm® SA8775P	1.381 ms	0 - 48 MB	NPU
Midas-V2	TFLITE	w8a8	Qualcomm® QCS9075	1.193 ms	0 - 20 MB	NPU
Midas-V2	TFLITE	w8a8	Qualcomm® QCM6690	8.35 ms	0 - 171 MB	NPU
Midas-V2	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	1.569 ms	0 - 78 MB	NPU
Midas-V2	TFLITE	w8a8	Qualcomm® SA7255P	2.486 ms	0 - 44 MB	NPU
Midas-V2	TFLITE	w8a8	Qualcomm® SA8295P	2.013 ms	0 - 43 MB	NPU
Midas-V2	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.602 ms	0 - 50 MB	NPU
Midas-V2	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	1.351 ms	0 - 167 MB	NPU

License

• The license for the original implementation of Midas-V2 can be found here.

References

• Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer
• 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.