library_name: pytorch
license: other
tags:
- android
pipeline_tag: depth-estimation
Depth-Anything-V3: Optimized for Qualcomm Devices
Depth Anything 3 (DA3), a model that predicts spatially consistent geometry from arbitrary visual inputs, with or without known camera poses.
This is based on the implementation of Depth-Anything-V3 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 2.37, ONNX Runtime 1.23.0 | Download |
| QNN_DLC | float | Universal | QAIRT 2.42 | Download |
| TFLITE | float | Universal | QAIRT 2.42, TFLite 2.17.0 | Download |
For more device-specific assets and performance metrics, visit Depth-Anything-V3 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 Depth-Anything-V3 on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.depth_estimation
Model Stats:
- Model checkpoint: da3-small
- Input resolution: 518x518
- Number of parameters: 24.7M
- Model size (float): 94.3 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| Depth-Anything-V3 | ONNX | float | Snapdragon® X Elite | 81.022 ms | 83 - 83 MB | NPU |
| Depth-Anything-V3 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 69.342 ms | 0 - 861 MB | NPU |
| Depth-Anything-V3 | ONNX | float | Qualcomm® QCS8550 (Proxy) | 89.491 ms | 0 - 96 MB | NPU |
| Depth-Anything-V3 | ONNX | float | Qualcomm® QCS9075 | 94.986 ms | 3 - 9 MB | NPU |
| Depth-Anything-V3 | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 51.232 ms | 2 - 698 MB | NPU |
| Depth-Anything-V3 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 45.786 ms | 1 - 686 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Snapdragon® X Elite | 78.46 ms | 3 - 3 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 55.146 ms | 3 - 748 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 177.187 ms | 2 - 566 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 77.211 ms | 3 - 5 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Qualcomm® SA8775P | 370.381 ms | 1 - 567 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Qualcomm® QCS9075 | 108.358 ms | 3 - 9 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 135.218 ms | 3 - 772 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Qualcomm® SA7255P | 177.187 ms | 2 - 566 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Qualcomm® SA8295P | 120.007 ms | 0 - 562 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 40.486 ms | 0 - 581 MB | NPU |
| Depth-Anything-V3 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 35.151 ms | 3 - 599 MB | NPU |
| Depth-Anything-V3 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 47.829 ms | 1 - 694 MB | NPU |
| Depth-Anything-V3 | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 163.259 ms | 1 - 539 MB | NPU |
| Depth-Anything-V3 | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 66.84 ms | 1 - 11 MB | NPU |
| Depth-Anything-V3 | TFLITE | float | Qualcomm® SA8775P | 73.83 ms | 1 - 539 MB | NPU |
| Depth-Anything-V3 | TFLITE | float | Qualcomm® QCS9075 | 86.4 ms | 1 - 82 MB | NPU |
| Depth-Anything-V3 | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 103.794 ms | 1 - 699 MB | NPU |
| Depth-Anything-V3 | TFLITE | float | Qualcomm® SA7255P | 163.259 ms | 1 - 539 MB | NPU |
| Depth-Anything-V3 | TFLITE | float | Qualcomm® SA8295P | 101.151 ms | 1 - 528 MB | NPU |
| Depth-Anything-V3 | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 34.202 ms | 0 - 541 MB | NPU |
| Depth-Anything-V3 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 30.256 ms | 1 - 545 MB | NPU |
License
- The license for the original implementation of Depth-Anything-V3 can be found here.
References
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.