Aggregated Contextual Transformations for High-Resolution Image Inpainting
Paper • 2104.01431 • Published
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 2.50, ONNX Runtime 1.30.0 | Download |
| QNN_DLC | float | Universal | QAIRT 2.50 | Download |
| TFLITE | float | Universal | QAIRT 2.50 | 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:
- Input resolution: 512x512
- Model checkpoint: CelebAHQ
- Model size (float): 58.0 MB
- Number of parameters: 15.2M
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| AOT-GAN | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 44.518 ms | 5 - 468 MB | NPU |
| AOT-GAN | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 69.562 ms | 9 - 545 MB | NPU |
| AOT-GAN | ONNX | float | Snapdragon® X2 Elite | 47.605 ms | 8 - 8 MB | NPU |
| AOT-GAN | ONNX | float | Snapdragon® X Elite | 124.497 ms | 32 - 32 MB | NPU |
| AOT-GAN | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 87.764 ms | 0 - 645 MB | NPU |
| AOT-GAN | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 180.904 ms | 1 - 577 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 176.645 ms | 4 - 12 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 120.749 ms | 0 - 38 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® QCS8450 | 180.904 ms | 1 - 577 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 161.891 ms | 4 - 11 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 124.497 ms | 32 - 32 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 69.562 ms | 9 - 545 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 44.782 ms | 3 - 454 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 67.189 ms | 1 - 532 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® X2 Elite | 45.805 ms | 4 - 4 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® X Elite | 119.081 ms | 4 - 4 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 85.525 ms | 3 - 649 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 181.775 ms | 4 - 576 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 173.05 ms | 4 - 14 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 117.754 ms | 4 - 7 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® SA8775P | 158.198 ms | 1 - 502 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® SA8650P | 158.198 ms | 1 - 502 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® SA8255P | 158.198 ms | 1 - 502 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® QCS8450 | 181.775 ms | 4 - 576 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 158.14 ms | 4 - 13 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 119.081 ms | 4 - 4 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 67.189 ms | 1 - 532 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® SA8295P | 181.071 ms | 1 - 441 MB | NPU |
| AOT-GAN | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 44.902 ms | 2 - 465 MB | NPU |
| AOT-GAN | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 67.832 ms | 3 - 542 MB | NPU |
| AOT-GAN | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 85.94 ms | 3 - 677 MB | NPU |
| AOT-GAN | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 181.427 ms | 4 - 607 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 172.81 ms | 3 - 48 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 118.608 ms | 3 - 6 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® SA8775P | 158.41 ms | 3 - 514 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® SA8650P | 158.41 ms | 3 - 514 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® SA8255P | 158.41 ms | 3 - 514 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® QCS8450 | 181.427 ms | 4 - 607 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 157.318 ms | 2 - 46 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 67.832 ms | 3 - 542 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® SA8295P | 181.844 ms | 3 - 456 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.