README.md

---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: other

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/centerpoint/web-assets/model_demo.png)

# CenterPoint: Optimized for Qualcomm Devices

CenterPoint is a LiDAR-based 3D object detection model that detects objects by predicting their centers and regressing other attributes. It is designed for high accuracy and real-time performance in autonomous driving applications.

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/v0.59.0/src/qai_hub_models/models/centerpoint) 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 |
|---|---|---|---|---|
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/centerpoint/releases/v0.59.0/centerpoint-qnn_dlc-float.zip)
| TFLITE | float | Universal |  | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/centerpoint/releases/v0.59.0/centerpoint-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[CenterPoint on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/centerpoint)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/centerpoint) 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 [CenterPoint on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/centerpoint) for usage instructions.

## Model Details

**Model Type:** Model_use_case.driver_assistance

**Model Stats:**
- Model checkpoint: PointPillars
- Input resolution: 5x20x5, 5x4, 5
- Number of parameters: 21.8M
- Model size: 83.3 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| CenterPoint | QNN_DLC | float | Snapdragon® X2 Elite | 185.015 ms | 2 - 2 MB | NPU
| CenterPoint | QNN_DLC | float | Snapdragon® X Elite | 326.336 ms | 2 - 2 MB | NPU
| CenterPoint | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 249.841 ms | 0 - 750 MB | NPU
| CenterPoint | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 521.899 ms | 2 - 737 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 920.314 ms | 1 - 450 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 332.766 ms | 2 - 6 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® SA8775P | 397.325 ms | 1 - 703 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® SA8650P | 397.325 ms | 1 - 703 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® SA8255P | 397.325 ms | 1 - 703 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® QCS8450 | 521.899 ms | 2 - 737 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 388.854 ms | 4 - 13 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 326.336 ms | 2 - 2 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 207.78 ms | 0 - 461 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® SA7255P | 920.314 ms | 1 - 450 MB | NPU
| CenterPoint | QNN_DLC | float | Qualcomm® SA8295P | 442.028 ms | 1 - 449 MB | NPU
| CenterPoint | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 207.78 ms | 0 - 461 MB | NPU
| CenterPoint | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 175.341 ms | 2 - 723 MB | NPU
| CenterPoint | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 4086.509 ms | 1835 - 1844 MB | CPU
| CenterPoint | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 5376.29 ms | 1860 - 1876 MB | CPU
| CenterPoint | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 6484.61 ms | 1847 - 1856 MB | CPU
| CenterPoint | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 5056.352 ms | 1893 - 1894 MB | CPU
| CenterPoint | TFLITE | float | Qualcomm® SA8775P | 5375.999 ms | 1810 - 1816 MB | CPU
| CenterPoint | TFLITE | float | Qualcomm® SA8650P | 5375.999 ms | 1810 - 1816 MB | CPU
| CenterPoint | TFLITE | float | Qualcomm® SA8255P | 5375.999 ms | 1810 - 1816 MB | CPU
| CenterPoint | TFLITE | float | Qualcomm® QCS8450 | 5376.29 ms | 1860 - 1876 MB | CPU
| CenterPoint | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 5153.011 ms | 2363 - 2385 MB | CPU
| CenterPoint | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 2647.479 ms | 1853 - 1865 MB | CPU
| CenterPoint | TFLITE | float | Qualcomm® SA7255P | 6484.61 ms | 1847 - 1856 MB | CPU
| CenterPoint | TFLITE | float | Qualcomm® SA8295P | 3446.214 ms | 1808 - 1814 MB | CPU
| CenterPoint | TFLITE | float | Snapdragon® 8 Elite Mobile | 2647.479 ms | 1853 - 1865 MB | CPU
| CenterPoint | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 803.09 ms | 1981 - 1991 MB | CPU

## License
* The license for the original implementation of CenterPoint can be found
  [here](https://github.com/tianweiy/CenterPoint/blob/master/LICENSE).



## 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).