SalsaNext: Optimized for Qualcomm Devices
SalsaNext is a LiDAR-based model designed for efficient and accurate semantic segmentation.
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 2.42, ONNX Runtime 1.24.1 | Download |
| ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | Download |
| QNN_DLC | float | Universal | QAIRT 2.43 | Download |
| QNN_DLC | w8a16 | Universal | QAIRT 2.43 | Download |
| TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | Download |
For more device-specific assets and performance metrics, visit SalsaNext 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 SalsaNext on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.semantic_segmentation
Model Stats:
- Model checkpoint: SalsaNext
- Input resolution: 1x5x64x2048
- Number of parameters: 6.71M
- Model size (float): 25.7 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| SalsaNext | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.587 ms | 14 - 204 MB | NPU |
| SalsaNext | ONNX | float | Snapdragon® X2 Elite | 15.046 ms | 33 - 33 MB | NPU |
| SalsaNext | ONNX | float | Snapdragon® X Elite | 32.536 ms | 33 - 33 MB | NPU |
| SalsaNext | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 25.689 ms | 25 - 294 MB | NPU |
| SalsaNext | ONNX | float | Qualcomm® QCS8550 (Proxy) | 33.869 ms | 23 - 25 MB | NPU |
| SalsaNext | ONNX | float | Qualcomm® QCS9075 | 40.203 ms | 24 - 27 MB | NPU |
| SalsaNext | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.598 ms | 23 - 212 MB | NPU |
| SalsaNext | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 18.513 ms | 13 - 209 MB | NPU |
| SalsaNext | ONNX | w8a16 | Snapdragon® X2 Elite | 16.597 ms | 16 - 16 MB | NPU |
| SalsaNext | ONNX | w8a16 | Snapdragon® X Elite | 34.38 ms | 18 - 18 MB | NPU |
| SalsaNext | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 29.374 ms | 15 - 304 MB | NPU |
| SalsaNext | ONNX | w8a16 | Qualcomm® QCS6490 | 3794.996 ms | 357 - 361 MB | CPU |
| SalsaNext | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 36.105 ms | 10 - 22 MB | NPU |
| SalsaNext | ONNX | w8a16 | Qualcomm® QCS9075 | 34.234 ms | 14 - 17 MB | NPU |
| SalsaNext | ONNX | w8a16 | Qualcomm® QCM6690 | 2029.684 ms | 327 - 336 MB | CPU |
| SalsaNext | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 24.348 ms | 12 - 211 MB | NPU |
| SalsaNext | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1829.089 ms | 389 - 398 MB | CPU |
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.371 ms | 3 - 191 MB | NPU |
| SalsaNext | QNN_DLC | float | Snapdragon® X2 Elite | 14.969 ms | 3 - 3 MB | NPU |
| SalsaNext | QNN_DLC | float | Snapdragon® X Elite | 30.163 ms | 3 - 3 MB | NPU |
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 22.362 ms | 0 - 261 MB | NPU |
| SalsaNext | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 117.213 ms | 1 - 177 MB | NPU |
| SalsaNext | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 30.523 ms | 3 - 63 MB | NPU |
| SalsaNext | QNN_DLC | float | Qualcomm® QCS9075 | 40.093 ms | 3 - 17 MB | NPU |
| SalsaNext | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 54.675 ms | 0 - 295 MB | NPU |
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 21.137 ms | 3 - 187 MB | NPU |
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 19.798 ms | 1 - 314 MB | NPU |
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 18.42 ms | 1 - 1 MB | NPU |
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® X Elite | 43.732 ms | 1 - 1 MB | NPU |
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 34.125 ms | 1 - 368 MB | NPU |
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 76.99 ms | 1 - 257 MB | NPU |
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 45.505 ms | 1 - 3 MB | NPU |
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 49.855 ms | 3 - 11 MB | NPU |
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 84.019 ms | 1 - 423 MB | NPU |
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 28.605 ms | 1 - 276 MB | NPU |
| SalsaNext | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.369 ms | 10 - 198 MB | NPU |
| SalsaNext | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 22.213 ms | 0 - 265 MB | NPU |
| SalsaNext | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 116.748 ms | 10 - 188 MB | NPU |
| SalsaNext | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 30.621 ms | 6 - 8 MB | NPU |
| SalsaNext | TFLITE | float | Qualcomm® QCS9075 | 39.69 ms | 10 - 38 MB | NPU |
| SalsaNext | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 55.919 ms | 0 - 302 MB | NPU |
| SalsaNext | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.403 ms | 9 - 195 MB | NPU |
License
- The license for the original implementation of SalsaNext can be found here.
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.