README.md

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

---

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

# StereoNet: Optimized for Qualcomm Devices

StereoNet is an end-to-end deep architecture for real-time stereo matching that produces high-quality, edge-preserved disparity maps from a rectified stereo image pair.

This is based on the implementation of StereoNet found [here](https://github.com/andrewlstewart/StereoNet_PyTorch).
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/stereonet) 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 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/stereonet/releases/v0.50.2/stereonet-onnx-float.zip)
| QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/stereonet/releases/v0.50.2/stereonet-qnn_dlc-float.zip)
| TFLITE | float | Universal | QAIRT 2.43, TFLite 2.19.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/stereonet/releases/v0.50.2/stereonet-tflite-float.zip)

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


### 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/stereonet) 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 [StereoNet on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/stereonet) for usage instructions.

## Model Details

**Model Type:** Model_use_case.depth_estimation

**Model Stats:**
- Model checkpoint: KeystoneDepth (epoch=21-step=696366.ckpt)
- Input resolution: 786x490
- Number of parameters: 1.94M
- Model size (float): 7.41 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| StereoNet | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 184.723 ms | 6 - 1359 MB | NPU
| StereoNet | ONNX | float | Snapdragon® X2 Elite | 366.346 ms | 20 - 20 MB | NPU
| StereoNet | ONNX | float | Snapdragon® X Elite | 331.059 ms | 19 - 19 MB | NPU
| StereoNet | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 261.303 ms | 6 - 1981 MB | NPU
| StereoNet | ONNX | float | Qualcomm® QCS8550 (Proxy) | 392.531 ms | 0 - 25 MB | NPU
| StereoNet | ONNX | float | Qualcomm® QCS9075 | 513.376 ms | 3 - 6 MB | NPU
| StereoNet | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 218.871 ms | 3 - 1320 MB | NPU
| StereoNet | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 173.6 ms | 3 - 1358 MB | NPU
| StereoNet | QNN_DLC | float | Snapdragon® X2 Elite | 162.354 ms | 3 - 3 MB | NPU
| StereoNet | QNN_DLC | float | Snapdragon® X Elite | 313.62 ms | 3 - 3 MB | NPU
| StereoNet | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 247.929 ms | 3 - 1976 MB | NPU
| StereoNet | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 1207.787 ms | 1 - 1349 MB | NPU
| StereoNet | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 374.062 ms | 3 - 6 MB | NPU
| StereoNet | QNN_DLC | float | Qualcomm® SA8775P | 404.265 ms | 1 - 1348 MB | NPU
| StereoNet | QNN_DLC | float | Qualcomm® QCS9075 | 491.28 ms | 3 - 9 MB | NPU
| StereoNet | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 725.806 ms | 3 - 2152 MB | NPU
| StereoNet | QNN_DLC | float | Qualcomm® SA7255P | 1207.787 ms | 1 - 1349 MB | NPU
| StereoNet | QNN_DLC | float | Qualcomm® SA8295P | 474.346 ms | 0 - 1499 MB | NPU
| StereoNet | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 203.368 ms | 3 - 1334 MB | NPU
| StereoNet | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 238.584 ms | 72 - 1691 MB | NPU
| StereoNet | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 309.589 ms | 72 - 2206 MB | NPU
| StereoNet | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 1322.818 ms | 74 - 1629 MB | NPU
| StereoNet | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 442.285 ms | 74 - 78 MB | NPU
| StereoNet | TFLITE | float | Qualcomm® SA8775P | 528.953 ms | 74 - 1629 MB | NPU
| StereoNet | TFLITE | float | Qualcomm® QCS9075 | 639.633 ms | 72 - 177 MB | NPU
| StereoNet | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 758.738 ms | 75 - 2482 MB | NPU
| StereoNet | TFLITE | float | Qualcomm® SA7255P | 1322.818 ms | 74 - 1629 MB | NPU
| StereoNet | TFLITE | float | Qualcomm® SA8295P | 562.268 ms | 74 - 1726 MB | NPU
| StereoNet | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 256.504 ms | 52 - 1639 MB | NPU

## License
* The license for the original implementation of StereoNet can be found
  [here](https://github.com/andrewlstewart/StereoNet_PyTorch/blob/main/LICENSE).

## References
* [StereoNet: Guided Hierarchical Refinement for Real-Time Edge-Aware Depth Prediction](https://arxiv.org/abs/1807.08865)
* [Source Model Implementation](https://github.com/andrewlstewart/StereoNet_PyTorch)

## 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).