Hugging Face's logo

qualcomm
/
SuperPoint

Keypoint Detection
PyTorch
real_time
android
Model card Files
xet
 Community
SuperPoint / README.md
qaihm-bot's picture
qaihm-bot
v0.66.0
9a90f68 verified
preview code
|
Raw History Blame Contribute Delete
13.3 kB
---
library_name: pytorch
license: mit
tags:
- real_time
- android
pipeline_tag: keypoint-detection
---
![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/superpoint/web-assets/model_demo.png)
# SuperPoint: Optimized for Qualcomm Devices
SuperPoint is a self-supervised framework for detecting interest points and computing their descriptors in a single forward pass. It uses a VGG-style encoder shared between a keypoint detector head and a descriptor head, trained via homographic adaptation on synthetic shapes and real images. The model outputs up to 500 keypoints per image with 256-dimensional L2-normalised descriptors, suitable for feature matching, visual localisation, and SLAM.
This is based on the implementation of SuperPoint found [here](https://github.com/rpautrat/SuperPoint).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.66.0/src/qai_hub_models/models/superpoint) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) 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](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/superpoint/releases/v0.66.0/superpoint-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.50, ONNX Runtime 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/superpoint/releases/v0.66.0/superpoint-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/superpoint/releases/v0.66.0/superpoint-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/superpoint/releases/v0.66.0/superpoint-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/superpoint/releases/v0.66.0/superpoint-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/superpoint/releases/v0.66.0/superpoint-tflite-w8a8.zip)
For more device-specific assets and performance metrics, visit **[SuperPoint on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/superpoint)**.
### Option 2: Export with Custom Configurations
Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.66.0/src/qai_hub_models/models/superpoint) 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 [SuperPoint on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.66.0/src/qai_hub_models/models/superpoint) for usage instructions.
## Model Details
**Model Type:** Model_use_case.pose_estimation
**Model Stats:**
- Descriptor dimension: 256
- Input resolution: 480x640
- Model checkpoint: superpoint_v6_from_tf.pth
- Model size (float): 19.6 MB
- Model size (w8a8): 3.22 MB
- Number of parameters: 5.12M
- Output keypoints (max): 500
## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| SuperPoint | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 6.878 ms | 28 - 196 MB | NPU
| SuperPoint | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 9.69 ms | 27 - 196 MB | NPU
| SuperPoint | ONNX | float | Snapdragon® X2 Elite | 7.545 ms | 40 - 40 MB | NPU
| SuperPoint | ONNX | float | Snapdragon® X Elite | 17.115 ms | 40 - 40 MB | NPU
| SuperPoint | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 28.558 ms | 19 - 259 MB | NPU
| SuperPoint | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 28.595 ms | 7 - 18 MB | NPU
| SuperPoint | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 16.91 ms | 23 - 29 MB | NPU
| SuperPoint | ONNX | float | Qualcomm® QCS8450 | 28.558 ms | 19 - 259 MB | NPU
| SuperPoint | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 17.115 ms | 40 - 40 MB | NPU
| SuperPoint | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 9.69 ms | 27 - 196 MB | NPU
| SuperPoint | ONNX | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 5.498 ms | 34 - 202 MB | NPU
| SuperPoint | ONNX | float | Snapdragon® 8 Elite Gen 6 Mobile | 6.594 ms | 33 - 201 MB | NPU
| SuperPoint | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 6.516 ms | 6 - 132 MB | NPU
| SuperPoint | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 8.295 ms | 8 - 123 MB | NPU
| SuperPoint | ONNX | w8a8 | Snapdragon® X2 Elite | 6.783 ms | 8 - 8 MB | NPU
| SuperPoint | ONNX | w8a8 | Snapdragon® X Elite | 20.927 ms | 7 - 7 MB | NPU
| SuperPoint | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 24.54 ms | 8 - 161 MB | NPU
| SuperPoint | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 19.575 ms | 7 - 11 MB | NPU
| SuperPoint | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 19.988 ms | 7 - 167 MB | NPU
| SuperPoint | ONNX | w8a8 | Qualcomm® QCS8450 | 24.54 ms | 8 - 161 MB | NPU
| SuperPoint | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 20.756 ms | 7 - 10 MB | NPU
| SuperPoint | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 20.927 ms | 7 - 7 MB | NPU
| SuperPoint | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 8.295 ms | 8 - 123 MB | NPU
| SuperPoint | ONNX | w8a8 | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 5.39 ms | 8 - 137 MB | NPU
| SuperPoint | ONNX | w8a8 | Snapdragon® 8 Elite Gen 6 Mobile | 10.191 ms | 8 - 119 MB | NPU
| SuperPoint | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 6.767 ms | 4 - 168 MB | NPU
| SuperPoint | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 9.605 ms | 4 - 163 MB | NPU
| SuperPoint | QNN_DLC | float | Snapdragon® X2 Elite | 7.936 ms | 7 - 7 MB | NPU
| SuperPoint | QNN_DLC | float | Snapdragon® X Elite | 16.54 ms | 7 - 7 MB | NPU
| SuperPoint | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 30.848 ms | 4 - 231 MB | NPU
| SuperPoint | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 27.41 ms | 7 - 29 MB | NPU
| SuperPoint | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 16.13 ms | 7 - 9 MB | NPU
| SuperPoint | QNN_DLC | float | Qualcomm® SA8775P | 23.662 ms | 0 - 160 MB | NPU
| SuperPoint | QNN_DLC | float | Qualcomm® SA8650P | 23.662 ms | 0 - 160 MB | NPU
| SuperPoint | QNN_DLC | float | Qualcomm® SA8255P | 23.662 ms | 0 - 160 MB | NPU
| SuperPoint | QNN_DLC | float | Qualcomm® QCS8450 | 30.848 ms | 4 - 231 MB | NPU
| SuperPoint | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 24.532 ms | 7 - 28 MB | NPU
| SuperPoint | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 16.54 ms | 7 - 7 MB | NPU
| SuperPoint | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 9.605 ms | 4 - 163 MB | NPU
| SuperPoint | QNN_DLC | float | Qualcomm® SA8295P | 31.192 ms | 0 - 175 MB | NPU
| SuperPoint | QNN_DLC | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 6.152 ms | 4 - 168 MB | NPU
| SuperPoint | QNN_DLC | float | Snapdragon® 8 Elite Gen 6 Mobile | 6.594 ms | 7 - 173 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 6.353 ms | 2 - 124 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 7.928 ms | 1 - 114 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 6.994 ms | 2 - 2 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Snapdragon® X Elite | 20.752 ms | 2 - 2 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 23.892 ms | 2 - 152 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 18.848 ms | 1 - 9 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 19.597 ms | 2 - 4 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Qualcomm® SA8775P | 19.86 ms | 2 - 105 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Qualcomm® SA8650P | 19.86 ms | 2 - 105 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Qualcomm® SA8255P | 19.86 ms | 2 - 105 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 23.892 ms | 2 - 152 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 20.248 ms | 3 - 10 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 20.752 ms | 2 - 2 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 7.928 ms | 1 - 114 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Qualcomm® SA8295P | 26.262 ms | 2 - 109 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 5.29 ms | 1 - 127 MB | NPU
| SuperPoint | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 6 Mobile | 9.909 ms | 2 - 113 MB | NPU
| SuperPoint | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 6.868 ms | 9 - 178 MB | NPU
| SuperPoint | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 9.311 ms | 10 - 182 MB | NPU
| SuperPoint | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 30.481 ms | 12 - 247 MB | NPU
| SuperPoint | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 28.578 ms | 11 - 36 MB | NPU
| SuperPoint | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 16.746 ms | 12 - 15 MB | NPU
| SuperPoint | TFLITE | float | Qualcomm® SA8775P | 24.709 ms | 12 - 176 MB | NPU
| SuperPoint | TFLITE | float | Qualcomm® SA8650P | 24.709 ms | 12 - 176 MB | NPU
| SuperPoint | TFLITE | float | Qualcomm® SA8255P | 24.709 ms | 12 - 176 MB | NPU
| SuperPoint | TFLITE | float | Qualcomm® QCS8450 | 30.481 ms | 12 - 247 MB | NPU
| SuperPoint | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 25.707 ms | 0 - 24 MB | NPU
| SuperPoint | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 9.311 ms | 10 - 182 MB | NPU
| SuperPoint | TFLITE | float | Qualcomm® SA7255P | 86.98 ms | 12 - 175 MB | NPU
| SuperPoint | TFLITE | float | Qualcomm® SA8295P | 28.997 ms | 12 - 191 MB | NPU
| SuperPoint | TFLITE | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 5.249 ms | 1 - 170 MB | NPU
| SuperPoint | TFLITE | float | Snapdragon® 8 Elite Gen 6 Mobile | 6.5 ms | 10 - 184 MB | NPU
| SuperPoint | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 6.646 ms | 2 - 125 MB | NPU
| SuperPoint | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 8.403 ms | 0 - 113 MB | NPU
| SuperPoint | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 24.822 ms | 3 - 154 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 580.219 ms | 299 - 686 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 19.277 ms | 3 - 13 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 20.207 ms | 3 - 5 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® SA8775P | 20.427 ms | 3 - 105 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® SA8650P | 20.427 ms | 3 - 105 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® SA8255P | 20.427 ms | 3 - 105 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® QCS8450 | 24.822 ms | 3 - 154 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 20.935 ms | 2 - 12 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 702.379 ms | 260 - 458 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 204.681 ms | 299 - 474 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 8.403 ms | 0 - 113 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® SA7255P | 39.934 ms | 3 - 104 MB | NPU
| SuperPoint | TFLITE | w8a8 | Qualcomm® SA8295P | 27.183 ms | 3 - 108 MB | NPU
| SuperPoint | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 204.681 ms | 299 - 474 MB | NPU
| SuperPoint | TFLITE | w8a8 | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 5.45 ms | 2 - 128 MB | NPU
| SuperPoint | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 6 Mobile | 10.277 ms | 2 - 114 MB | NPU
## License
* The license for the original implementation of SuperPoint can be found
[here](https://github.com/rpautrat/SuperPoint/blob/master/LICENSE.txt).
## References
* [SuperPoint: Self-Supervised Interest Point Detection and Description](https://arxiv.org/abs/1712.07629)
* [Source Model Implementation](https://github.com/rpautrat/SuperPoint)
## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).