AG-Pose: Optimized for Qualcomm Devices
AG-Pose is a category-level 6D object pose estimation model that predicts rotation, translation, and size of objects from RGB-D input. It uses DINOv2-S (ViT-S/14) as the RGB backbone and PointNet++ for 3D point cloud processing, with an Instance-Adaptive Keypoint Detector (IAKD) and Geometric-Aware Feature Aggregator (GAFA) for robust pose estimation across 6 object categories.
This is based on the implementation of AG-Pose 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 2.51, ONNX Runtime 1.30.0 | Download |
| QNN_DLC | float | Universal | QAIRT 2.51 | Download |
For more device-specific assets and performance metrics, visit AG-Pose 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 AG-Pose on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.pose_estimation
Model Stats:
- Model checkpoint: epoch_30.pt
- Input resolution: 1 x 3 x 224 x 224
- Number of parameters: 30.4M
- Model size (float): ~116 MB
- Supported categories: bottle, bowl, camera, can, laptop, mug
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| backbone | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 47.734 ms | 0 - 369 MB | NPU |
| backbone | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 49.579 ms | 1 - 349 MB | NPU |
| backbone | ONNX | float | Snapdragon® X2 Elite | 52.593 ms | 2 - 2 MB | NPU |
| backbone | ONNX | float | Snapdragon® X Elite | 86.389 ms | 63 - 63 MB | NPU |
| backbone | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 86.339 ms | 1 - 453 MB | NPU |
| backbone | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 80.142 ms | 0 - 5 MB | NPU |
| backbone | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 83.698 ms | 1 - 4 MB | NPU |
| backbone | ONNX | float | Qualcomm® QCS8450 | 86.339 ms | 1 - 453 MB | NPU |
| backbone | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 87.228 ms | 0 - 4 MB | NPU |
| backbone | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 86.389 ms | 63 - 63 MB | NPU |
| backbone | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 49.579 ms | 1 - 349 MB | NPU |
| backbone | ONNX | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 44.305 ms | 1 - 343 MB | NPU |
| backbone | ONNX | float | Snapdragon® 8 Elite Gen 6 Mobile | 47.804 ms | 1 - 336 MB | NPU |
| backbone | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 48.899 ms | 0 - 418 MB | NPU |
| backbone | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 50.054 ms | 1 - 395 MB | NPU |
| backbone | QNN_DLC | float | Snapdragon® X2 Elite | 53.174 ms | 1 - 1 MB | NPU |
| backbone | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 89.184 ms | 0 - 511 MB | NPU |
| backbone | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 80.841 ms | 1 - 4 MB | NPU |
| backbone | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 83.162 ms | 1 - 326 MB | NPU |
| backbone | QNN_DLC | float | Qualcomm® QCS8450 | 89.184 ms | 0 - 511 MB | NPU |
| backbone | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 88.369 ms | 1 - 4 MB | NPU |
| backbone | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 50.054 ms | 1 - 395 MB | NPU |
| backbone | QNN_DLC | float | Qualcomm® SA8295P | 96.186 ms | 1 - 392 MB | NPU |
| backbone | QNN_DLC | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 46.475 ms | 1 - 401 MB | NPU |
| backbone | QNN_DLC | float | Snapdragon® 8 Elite Gen 6 Mobile | 48.619 ms | 1 - 396 MB | NPU |
| rotation_head | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.148 ms | 0 - 25 MB | NPU |
| rotation_head | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.151 ms | 0 - 29 MB | NPU |
| rotation_head | ONNX | float | Snapdragon® X2 Elite | 0.137 ms | 1 - 1 MB | NPU |
| rotation_head | ONNX | float | Snapdragon® X Elite | 0.219 ms | 0 - 0 MB | NPU |
| rotation_head | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 0.292 ms | 0 - 42 MB | NPU |
| rotation_head | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 0.248 ms | 0 - 4 MB | NPU |
| rotation_head | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.231 ms | 0 - 2 MB | NPU |
| rotation_head | ONNX | float | Qualcomm® QCS8450 | 0.292 ms | 0 - 42 MB | NPU |
| rotation_head | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 0.26 ms | 0 - 4 MB | NPU |
| rotation_head | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 0.219 ms | 0 - 0 MB | NPU |
| rotation_head | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 0.151 ms | 0 - 29 MB | NPU |
| rotation_head | ONNX | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 0.128 ms | 0 - 26 MB | NPU |
| rotation_head | ONNX | float | Snapdragon® 8 Elite Gen 6 Mobile | 0.134 ms | 0 - 26 MB | NPU |
| rotation_head | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.144 ms | 0 - 25 MB | NPU |
| rotation_head | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.144 ms | 0 - 29 MB | NPU |
| rotation_head | QNN_DLC | float | Snapdragon® X2 Elite | 0.226 ms | 0 - 0 MB | NPU |
| rotation_head | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 0.279 ms | 0 - 40 MB | NPU |
| rotation_head | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 0.238 ms | 0 - 3 MB | NPU |
| rotation_head | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.225 ms | 0 - 2 MB | NPU |
| rotation_head | QNN_DLC | float | Qualcomm® QCS8450 | 0.279 ms | 0 - 40 MB | NPU |
| rotation_head | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 0.261 ms | 0 - 3 MB | NPU |
| rotation_head | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 0.144 ms | 0 - 29 MB | NPU |
| rotation_head | QNN_DLC | float | Qualcomm® SA8295P | 0.423 ms | 0 - 23 MB | NPU |
| rotation_head | QNN_DLC | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 0.13 ms | 0 - 26 MB | NPU |
| rotation_head | QNN_DLC | float | Snapdragon® 8 Elite Gen 6 Mobile | 0.131 ms | 0 - 26 MB | NPU |
License
- The license for the original implementation of AG-Pose can be found here.
References
- AG-Pose: Instance-Adaptive and Geometric-Aware Keypoint Learning for Category-Level 6D Object Pose Estimation
- 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.
