AOT-GAN: Optimized for Qualcomm Devices

AOT-GAN is a machine learning model that allows to erase and in-paint part of given input image.

This is based on the implementation of AOT-GAN 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 1.24.1	Download
ONNX	w8a16	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
QNN_DLC	w8a16	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.17.0	Download

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

Model Details

Model Type: Model_use_case.image_editing

Model Stats:

• Model checkpoint: CelebAHQ
• Input resolution: 512x512
• Number of parameters: 15.2M
• Model size (float): 58.0 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
AOT-GAN	ONNX	float	Snapdragon® X Elite	149.561 ms	31 - 31 MB	NPU
AOT-GAN	ONNX	float	Snapdragon® 8 Gen 3 Mobile	102.267 ms	0 - 739 MB	NPU
AOT-GAN	ONNX	float	Qualcomm® QCS8550 (Proxy)	146.589 ms	0 - 457 MB	NPU
AOT-GAN	ONNX	float	Qualcomm® QCS9075	226.001 ms	4 - 11 MB	NPU
AOT-GAN	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	78.791 ms	6 - 633 MB	NPU
AOT-GAN	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	59.034 ms	10 - 504 MB	NPU
AOT-GAN	ONNX	float	Snapdragon® X2 Elite	61.048 ms	32 - 32 MB	NPU
AOT-GAN	ONNX	w8a16	Snapdragon® X Elite	85.697 ms	21 - 21 MB	NPU
AOT-GAN	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	61.863 ms	5 - 715 MB	NPU
AOT-GAN	ONNX	w8a16	Qualcomm® QCS6490	13192.303 ms	393 - 396 MB	CPU
AOT-GAN	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	82.146 ms	0 - 41 MB	NPU
AOT-GAN	ONNX	w8a16	Qualcomm® QCS9075	109.839 ms	5 - 8 MB	NPU
AOT-GAN	ONNX	w8a16	Qualcomm® QCM6690	6693.772 ms	328 - 336 MB	CPU
AOT-GAN	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	46.905 ms	5 - 527 MB	NPU
AOT-GAN	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	6586.752 ms	324 - 333 MB	CPU
AOT-GAN	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	34.755 ms	5 - 617 MB	NPU
AOT-GAN	ONNX	w8a16	Snapdragon® X2 Elite	37.831 ms	23 - 23 MB	NPU
AOT-GAN	QNN_DLC	float	Snapdragon® X Elite	125.33 ms	4 - 4 MB	NPU
AOT-GAN	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	88.875 ms	3 - 729 MB	NPU
AOT-GAN	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	544.1 ms	1 - 539 MB	NPU
AOT-GAN	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	123.911 ms	4 - 7 MB	NPU
AOT-GAN	QNN_DLC	float	Qualcomm® SA8775P	164.553 ms	1 - 541 MB	NPU
AOT-GAN	QNN_DLC	float	Qualcomm® QCS9075	213.201 ms	4 - 13 MB	NPU
AOT-GAN	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	206.059 ms	3 - 643 MB	NPU
AOT-GAN	QNN_DLC	float	Qualcomm® SA7255P	544.1 ms	1 - 539 MB	NPU
AOT-GAN	QNN_DLC	float	Qualcomm® SA8295P	179.664 ms	1 - 477 MB	NPU
AOT-GAN	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	71.016 ms	3 - 583 MB	NPU
AOT-GAN	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	47.017 ms	2 - 488 MB	NPU
AOT-GAN	QNN_DLC	float	Snapdragon® X2 Elite	52.376 ms	4 - 4 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Snapdragon® X Elite	84.389 ms	2 - 2 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	62.959 ms	2 - 722 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Qualcomm® QCS6490	298.385 ms	1 - 6 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	189.993 ms	2 - 510 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	82.726 ms	2 - 612 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Qualcomm® SA8775P	82.085 ms	2 - 512 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Qualcomm® QCS9075	110.079 ms	2 - 7 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Qualcomm® QCM6690	1286.762 ms	2 - 669 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Qualcomm® QCS8450 (Proxy)	144.598 ms	2 - 711 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Qualcomm® SA7255P	189.993 ms	2 - 510 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Qualcomm® SA8295P	115.664 ms	2 - 511 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	48.415 ms	2 - 563 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	124.262 ms	2 - 671 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	35.823 ms	2 - 631 MB	NPU
AOT-GAN	QNN_DLC	w8a16	Snapdragon® X2 Elite	38.732 ms	2 - 2 MB	NPU
AOT-GAN	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	89.569 ms	3 - 768 MB	NPU
AOT-GAN	TFLITE	float	Qualcomm® QCS8275 (Proxy)	544.416 ms	3 - 560 MB	NPU
AOT-GAN	TFLITE	float	Qualcomm® QCS8550 (Proxy)	123.17 ms	3 - 5 MB	NPU
AOT-GAN	TFLITE	float	Qualcomm® SA8775P	163.884 ms	0 - 557 MB	NPU
AOT-GAN	TFLITE	float	Qualcomm® QCS9075	210.807 ms	3 - 46 MB	NPU
AOT-GAN	TFLITE	float	Qualcomm® QCS8450 (Proxy)	200.48 ms	3 - 676 MB	NPU
AOT-GAN	TFLITE	float	Qualcomm® SA7255P	544.416 ms	3 - 560 MB	NPU
AOT-GAN	TFLITE	float	Qualcomm® SA8295P	179.019 ms	0 - 488 MB	NPU
AOT-GAN	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	70.922 ms	3 - 596 MB	NPU
AOT-GAN	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	46.972 ms	1 - 503 MB	NPU

License

• The license for the original implementation of AOT-GAN can be found here.

References

• Aggregated Contextual Transformations for High-Resolution Image Inpainting
• 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.