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 2.45, ONNX Runtime 1.25.0 | Download |
| QNN_DLC | float | Universal | QAIRT 2.45 | Download |
| TFLITE | float | Universal | QAIRT 2.45 | 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® X2 Elite | 54.777 ms | 209 - 209 MB | NPU |
| AOT-GAN | ONNX | float | Snapdragon® X Elite | 138.487 ms | 145 - 145 MB | NPU |
| AOT-GAN | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 95.572 ms | 11 - 710 MB | NPU |
| AOT-GAN | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 209.024 ms | 12 - 603 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® QCS8550 (Proxy) | 135.222 ms | 0 - 41 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® QCS8450 | 209.024 ms | 12 - 603 MB | NPU |
| AOT-GAN | ONNX | float | Snapdragon® 8 Elite Mobile | 75.266 ms | 7 - 618 MB | NPU |
| AOT-GAN | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 50.575 ms | 7 - 508 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® QCS9075 | 206.843 ms | 4 - 49 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® QCS8750 | 75.266 ms | 7 - 618 MB | NPU |
| AOT-GAN | ONNX | float | Qualcomm® QCS7181 | 138.487 ms | 145 - 145 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® X2 Elite | 50.761 ms | 4 - 4 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® X Elite | 122.511 ms | 4 - 4 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 88.652 ms | 0 - 690 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 198.648 ms | 3 - 606 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® QCS8275 | 540.878 ms | 1 - 543 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 119.192 ms | 4 - 7 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® QCS8450 | 198.648 ms | 3 - 606 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 69.573 ms | 1 - 572 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® SA8295P | 178.694 ms | 0 - 475 MB | NPU |
| AOT-GAN | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 46.406 ms | 3 - 490 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® SA7255P | 540.878 ms | 1 - 543 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® QCS9075 | 211.164 ms | 4 - 13 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® QCS8750 | 69.573 ms | 1 - 572 MB | NPU |
| AOT-GAN | QNN_DLC | float | Qualcomm® QCS7181 | 122.511 ms | 4 - 4 MB | NPU |
| AOT-GAN | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 88.7 ms | 3 - 726 MB | NPU |
| AOT-GAN | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 198.343 ms | 3 - 637 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® QCS8275 | 541.25 ms | 0 - 554 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 119.85 ms | 3 - 15 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® SA8775P | 2750.117 ms | 0 - 35 MB | GPU |
| AOT-GAN | TFLITE | float | Qualcomm® SA8650P | 2750.117 ms | 0 - 35 MB | GPU |
| AOT-GAN | TFLITE | float | Qualcomm® SA8255P | 2750.117 ms | 0 - 35 MB | GPU |
| AOT-GAN | TFLITE | float | Qualcomm® QCS8450 | 198.343 ms | 3 - 637 MB | NPU |
| AOT-GAN | TFLITE | float | Snapdragon® 8 Elite Mobile | 69.393 ms | 0 - 593 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® SA8295P | 178.583 ms | 3 - 494 MB | NPU |
| AOT-GAN | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 46.835 ms | 3 - 513 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® SA7255P | 541.25 ms | 0 - 554 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® QCS9075 | 210.932 ms | 2 - 45 MB | NPU |
| AOT-GAN | TFLITE | float | Qualcomm® QCS8750 | 69.393 ms | 0 - 593 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.