Deep Layer Aggregation
Paper • 1707.06484 • Published
DLA-102-X: Optimized for Qualcomm Devices
dla is a machine learning model that can classify images from the Imagenet dataset. It is a architecture which can be used as backbone.
This is based on the implementation of DLA-102-X 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 1.27.1 | Download |
| ONNX | w8a8 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | Download |
| QNN_DLC | float | Universal | QAIRT 2.45 | Download |
| QNN_DLC | w8a8 | 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 DLA-102-X 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 DLA-102-X on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.image_classification
Model Stats:
- Model checkpoint: Imagenet
- Input resolution: 224x224
- Number of parameters: 26.3M
- Model size (float): 100 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| DLA-102-X | ONNX | float | Snapdragon® X2 Elite | 1.324 ms | 2 - 2 MB | NPU |
| DLA-102-X | ONNX | float | Snapdragon® X Elite | 2.712 ms | 55 - 55 MB | NPU |
| DLA-102-X | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 1.874 ms | 0 - 154 MB | NPU |
| DLA-102-X | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 3.941 ms | 1 - 129 MB | NPU |
| DLA-102-X | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2.657 ms | 0 - 77 MB | NPU |
| DLA-102-X | ONNX | float | Qualcomm® QCS8450 | 3.941 ms | 1 - 129 MB | NPU |
| DLA-102-X | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 3.951 ms | 0 - 4 MB | NPU |
| DLA-102-X | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 2.712 ms | 55 - 55 MB | NPU |
| DLA-102-X | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 1.575 ms | 0 - 100 MB | NPU |
| DLA-102-X | ONNX | float | Snapdragon® 8 Elite Mobile | 1.575 ms | 0 - 100 MB | NPU |
| DLA-102-X | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.215 ms | 0 - 103 MB | NPU |
| DLA-102-X | ONNX | w8a8 | Snapdragon® X2 Elite | 0.712 ms | 1 - 1 MB | NPU |
| DLA-102-X | ONNX | w8a8 | Snapdragon® X Elite | 1.406 ms | 28 - 28 MB | NPU |
| DLA-102-X | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.986 ms | 0 - 152 MB | NPU |
| DLA-102-X | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 5.757 ms | 0 - 3 MB | NPU |
| DLA-102-X | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.398 ms | 0 - 30 MB | NPU |
| DLA-102-X | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.67 ms | 0 - 3 MB | NPU |
| DLA-102-X | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 1.406 ms | 28 - 28 MB | NPU |
| DLA-102-X | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.879 ms | 0 - 114 MB | NPU |
| DLA-102-X | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 0.879 ms | 0 - 114 MB | NPU |
| DLA-102-X | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.736 ms | 0 - 117 MB | NPU |
| DLA-102-X | QNN_DLC | float | Snapdragon® X2 Elite | 2.747 ms | 1 - 1 MB | NPU |
| DLA-102-X | QNN_DLC | float | Snapdragon® X Elite | 3.139 ms | 1 - 1 MB | NPU |
| DLA-102-X | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 2.004 ms | 1 - 149 MB | NPU |
| DLA-102-X | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 4.536 ms | 0 - 121 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 14.441 ms | 1 - 99 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2.837 ms | 1 - 3 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® SA8775P | 4.37 ms | 1 - 99 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® SA8650P | 4.37 ms | 1 - 99 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® SA8255P | 4.37 ms | 1 - 99 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® QCS8450 | 4.536 ms | 0 - 121 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 4.109 ms | 1 - 3 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 3.139 ms | 1 - 1 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 1.598 ms | 0 - 99 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® SA7255P | 14.441 ms | 1 - 99 MB | NPU |
| DLA-102-X | QNN_DLC | float | Qualcomm® SA8295P | 4.653 ms | 1 - 81 MB | NPU |
| DLA-102-X | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 1.598 ms | 0 - 99 MB | NPU |
| DLA-102-X | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.241 ms | 0 - 103 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.888 ms | 0 - 0 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Snapdragon® X Elite | 1.553 ms | 0 - 0 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.967 ms | 0 - 133 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 1.928 ms | 0 - 136 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 6.499 ms | 0 - 2 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8275 | 3.291 ms | 0 - 105 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.352 ms | 0 - 123 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® SA8775P | 1.811 ms | 0 - 107 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® SA8650P | 1.811 ms | 0 - 107 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® SA8255P | 1.811 ms | 0 - 107 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 1.928 ms | 0 - 136 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.604 ms | 0 - 2 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 1.553 ms | 0 - 0 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 10.623 ms | 0 - 234 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 1.703 ms | 0 - 103 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.788 ms | 0 - 106 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® SA7255P | 3.291 ms | 0 - 105 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Qualcomm® SA8295P | 2.16 ms | 0 - 108 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 0.788 ms | 0 - 106 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.619 ms | 0 - 107 MB | NPU |
| DLA-102-X | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.703 ms | 0 - 103 MB | NPU |
| DLA-102-X | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.987 ms | 0 - 193 MB | NPU |
| DLA-102-X | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 4.542 ms | 0 - 165 MB | NPU |
| DLA-102-X | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 14.381 ms | 0 - 145 MB | NPU |
| DLA-102-X | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2.747 ms | 0 - 3 MB | NPU |
| DLA-102-X | TFLITE | float | Qualcomm® SA8775P | 4.37 ms | 0 - 147 MB | NPU |
| DLA-102-X | TFLITE | float | Qualcomm® SA8650P | 4.37 ms | 0 - 147 MB | NPU |
| DLA-102-X | TFLITE | float | Qualcomm® SA8255P | 4.37 ms | 0 - 147 MB | NPU |
| DLA-102-X | TFLITE | float | Qualcomm® QCS8450 | 4.542 ms | 0 - 165 MB | NPU |
| DLA-102-X | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 4.085 ms | 0 - 57 MB | NPU |
| DLA-102-X | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 1.583 ms | 0 - 147 MB | NPU |
| DLA-102-X | TFLITE | float | Qualcomm® SA7255P | 14.381 ms | 0 - 145 MB | NPU |
| DLA-102-X | TFLITE | float | Qualcomm® SA8295P | 4.688 ms | 0 - 124 MB | NPU |
| DLA-102-X | TFLITE | float | Snapdragon® 8 Elite Mobile | 1.583 ms | 0 - 147 MB | NPU |
| DLA-102-X | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.238 ms | 0 - 153 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.828 ms | 0 - 134 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 1.719 ms | 0 - 135 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 5.587 ms | 0 - 29 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.138 ms | 0 - 4 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® SA8775P | 1.604 ms | 0 - 102 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® SA8650P | 1.604 ms | 0 - 102 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® SA8255P | 1.604 ms | 0 - 102 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® QCS8450 | 1.719 ms | 0 - 135 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.396 ms | 0 - 29 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 10.108 ms | 0 - 208 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 1.475 ms | 0 - 92 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.714 ms | 0 - 105 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Qualcomm® SA8295P | 1.982 ms | 0 - 93 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 0.714 ms | 0 - 105 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.57 ms | 0 - 95 MB | NPU |
| DLA-102-X | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.475 ms | 0 - 92 MB | NPU |
License
- The license for the original implementation of DLA-102-X 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.