README.md

---
library_name: pytorch
license: other
tags:
- bu_auto
- android
pipeline_tag: image-segmentation

---

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

# PointNet: Optimized for Qualcomm Devices

PointNet is a pioneering neural network architecture designed to directly consume unordered point cloud data for tasks such as classification and segmentation. It learns spatial features from raw 3D points without requiring voxelization or image projections.

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/main/src/qai_hub_models/models/pointnet) 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 2.45, ONNX Runtime 1.25.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pointnet/releases/v0.56.0/pointnet-onnx-float.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pointnet/releases/v0.56.0/pointnet-qnn_dlc-float.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pointnet/releases/v0.56.0/pointnet-tflite-float.zip)

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


### Option 2: Export with Custom Configurations

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

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

**Model Stats:**
- Model checkpoint: save
- Input resolution: 1x3x1024
- Model size: 13.2 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| PointNet | ONNX | float | Snapdragon® X2 Elite | 0.305 ms | 213 - 213 MB | NPU
| PointNet | ONNX | float | Snapdragon® X Elite | 0.655 ms | 149 - 149 MB | NPU
| PointNet | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.406 ms | 59 - 91 MB | NPU
| PointNet | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 0.865 ms | 0 - 33 MB | NPU
| PointNet | ONNX | float | Qualcomm® QCS8550 (Proxy) | 0.647 ms | 0 - 12 MB | NPU
| PointNet | ONNX | float | Qualcomm® QCS8450 | 0.865 ms | 0 - 33 MB | NPU
| PointNet | ONNX | float | Snapdragon® 8 Elite Mobile | 0.421 ms | 0 - 24 MB | NPU
| PointNet | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.322 ms | 0 - 19 MB | NPU
| PointNet | ONNX | float | Qualcomm® QCS9075 | 0.807 ms | 0 - 52 MB | NPU
| PointNet | ONNX | float | Qualcomm® QCS8750 | 0.421 ms | 0 - 24 MB | NPU
| PointNet | ONNX | float | Qualcomm® QCS7181 | 0.655 ms | 149 - 149 MB | NPU
| PointNet | QNN_DLC | float | Snapdragon® X2 Elite | 0.403 ms | 0 - 0 MB | NPU
| PointNet | QNN_DLC | float | Snapdragon® X Elite | 0.771 ms | 0 - 0 MB | NPU
| PointNet | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.41 ms | 0 - 42 MB | NPU
| PointNet | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 0.865 ms | 0 - 47 MB | NPU
| PointNet | QNN_DLC | float | Qualcomm® QCS8275 | 1.848 ms | 0 - 23 MB | NPU
| PointNet | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 0.659 ms | 0 - 67 MB | NPU
| PointNet | QNN_DLC | float | Qualcomm® QCS8450 | 0.865 ms | 0 - 47 MB | NPU
| PointNet | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 0.485 ms | 0 - 23 MB | NPU
| PointNet | QNN_DLC | float | Qualcomm® SA8295P | 1.119 ms | 0 - 20 MB | NPU
| PointNet | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.323 ms | 0 - 23 MB | NPU
| PointNet | QNN_DLC | float | Qualcomm® SA7255P | 1.848 ms | 0 - 23 MB | NPU
| PointNet | QNN_DLC | float | Qualcomm® QCS9075 | 0.799 ms | 0 - 2 MB | NPU
| PointNet | QNN_DLC | float | Qualcomm® QCS8750 | 0.485 ms | 0 - 23 MB | NPU
| PointNet | QNN_DLC | float | Qualcomm® QCS7181 | 0.771 ms | 0 - 0 MB | NPU
| PointNet | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.41 ms | 0 - 40 MB | NPU
| PointNet | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 0.861 ms | 0 - 46 MB | NPU
| PointNet | TFLITE | float | Qualcomm® QCS8275 | 1.885 ms | 0 - 24 MB | NPU
| PointNet | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 0.655 ms | 0 - 2 MB | NPU
| PointNet | TFLITE | float | Qualcomm® SA8775P | 4.276 ms | 0 - 16 MB | GPU
| PointNet | TFLITE | float | Qualcomm® SA8650P | 4.276 ms | 0 - 16 MB | GPU
| PointNet | TFLITE | float | Qualcomm® SA8255P | 4.276 ms | 0 - 16 MB | GPU
| PointNet | TFLITE | float | Qualcomm® QCS8450 | 0.861 ms | 0 - 46 MB | NPU
| PointNet | TFLITE | float | Snapdragon® 8 Elite Mobile | 0.483 ms | 0 - 24 MB | NPU
| PointNet | TFLITE | float | Qualcomm® SA8295P | 1.122 ms | 0 - 21 MB | NPU
| PointNet | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.324 ms | 0 - 24 MB | NPU
| PointNet | TFLITE | float | Qualcomm® SA7255P | 1.885 ms | 0 - 24 MB | NPU
| PointNet | TFLITE | float | Qualcomm® QCS9075 | 0.801 ms | 0 - 8 MB | NPU
| PointNet | TFLITE | float | Qualcomm® QCS8750 | 0.483 ms | 0 - 24 MB | NPU

## License
* The license for the original implementation of PointNet can be found
  [here](https://github.com/nikitakaraevv/pointnet/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).