--- 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/salsanext/web-assets/model_demo.png) # 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](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/salsanext) 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/salsanext/releases/v0.51.0/salsanext-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/salsanext/releases/v0.51.0/salsanext-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/salsanext/releases/v0.51.0/salsanext-tflite-float.zip) For more device-specific assets and performance metrics, visit **[SalsaNext on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/salsanext)**. ### 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/salsanext) 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](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/salsanext) 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 | 15.552 ms | 24 - 215 MB | NPU | SalsaNext | ONNX | float | Snapdragon® X2 Elite | 15.019 ms | 33 - 33 MB | NPU | SalsaNext | ONNX | float | Snapdragon® X Elite | 32.491 ms | 33 - 33 MB | NPU | SalsaNext | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 25.566 ms | 24 - 291 MB | NPU | SalsaNext | ONNX | float | Qualcomm® QCS8550 (Proxy) | 34.308 ms | 23 - 26 MB | NPU | SalsaNext | ONNX | float | Qualcomm® QCS9075 | 39.967 ms | 23 - 25 MB | NPU | SalsaNext | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.538 ms | 23 - 207 MB | NPU | SalsaNext | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 13.225 ms | 3 - 290 MB | NPU | SalsaNext | QNN_DLC | float | Snapdragon® X2 Elite | 13.855 ms | 3 - 3 MB | NPU | SalsaNext | QNN_DLC | float | Snapdragon® X Elite | 31.846 ms | 3 - 3 MB | NPU | SalsaNext | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 25.362 ms | 2 - 333 MB | NPU | SalsaNext | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 118.541 ms | 1 - 247 MB | NPU | SalsaNext | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 33.142 ms | 3 - 5 MB | NPU | SalsaNext | QNN_DLC | float | Qualcomm® QCS9075 | 37.152 ms | 3 - 17 MB | NPU | SalsaNext | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 65.734 ms | 2 - 346 MB | NPU | SalsaNext | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.532 ms | 2 - 269 MB | NPU | SalsaNext | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 13.192 ms | 10 - 296 MB | NPU | SalsaNext | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 25.707 ms | 7 - 341 MB | NPU | SalsaNext | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 118.704 ms | 10 - 257 MB | NPU | SalsaNext | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 33.655 ms | 5 - 7 MB | NPU | SalsaNext | TFLITE | float | Qualcomm® QCS9075 | 37.293 ms | 10 - 39 MB | NPU | SalsaNext | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 66.995 ms | 10 - 358 MB | NPU | SalsaNext | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.601 ms | 8 - 277 MB | NPU ## License * The license for the original implementation of SalsaNext can be found [here](https://github.com/TiagoCortinhal/SalsaNext/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).