--- 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 Mobile Deployment ## Semantic segmentation model optimized for LiDAR point cloud data SalsaNext is a LiDAR-based model designed for efficient and accurate semantic segmentation. This repository provides scripts to run SalsaNext on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/salsanext). ### 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 | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | SalsaNext | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 118.648 ms | 10 - 164 MB | NPU | [SalsaNext.tflite](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.tflite) | | SalsaNext | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 118.843 ms | 2 - 155 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.dlc) | | SalsaNext | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 60.783 ms | 10 - 285 MB | NPU | [SalsaNext.tflite](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.tflite) | | SalsaNext | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 60.893 ms | 3 - 269 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.dlc) | | SalsaNext | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 33.712 ms | 10 - 12 MB | NPU | [SalsaNext.tflite](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.tflite) | | SalsaNext | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 34.135 ms | 3 - 5 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.dlc) | | SalsaNext | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 34.804 ms | 20 - 33 MB | NPU | [SalsaNext.onnx.zip](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.onnx.zip) | | SalsaNext | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 39.143 ms | 10 - 165 MB | NPU | [SalsaNext.tflite](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.tflite) | | SalsaNext | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 162.128 ms | 2 - 155 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.dlc) | | SalsaNext | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 25.538 ms | 8 - 255 MB | NPU | [SalsaNext.tflite](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.tflite) | | SalsaNext | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 25.214 ms | 3 - 240 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.dlc) | | SalsaNext | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 26.361 ms | 25 - 240 MB | NPU | [SalsaNext.onnx.zip](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.onnx.zip) | | SalsaNext | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 21.024 ms | 9 - 164 MB | NPU | [SalsaNext.tflite](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.tflite) | | SalsaNext | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 20.671 ms | 3 - 161 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.dlc) | | SalsaNext | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 21.816 ms | 23 - 154 MB | NPU | [SalsaNext.onnx.zip](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.onnx.zip) | | SalsaNext | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | TFLITE | 15.752 ms | 10 - 169 MB | NPU | [SalsaNext.tflite](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.tflite) | | SalsaNext | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 15.523 ms | 3 - 161 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.dlc) | | SalsaNext | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | ONNX | 16.182 ms | 24 - 160 MB | NPU | [SalsaNext.onnx.zip](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.onnx.zip) | | SalsaNext | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 32.41 ms | 3 - 3 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.dlc) | | SalsaNext | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 33.248 ms | 33 - 33 MB | NPU | [SalsaNext.onnx.zip](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext.onnx.zip) | | SalsaNext | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 74.53 ms | 1 - 217 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext_w8a16.dlc) | | SalsaNext | w8a16 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 83.038 ms | 1 - 376 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext_w8a16.dlc) | | SalsaNext | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 40.53 ms | 1 - 4 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext_w8a16.dlc) | | SalsaNext | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 40.537 ms | 1 - 218 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext_w8a16.dlc) | | SalsaNext | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 30.535 ms | 1 - 323 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext_w8a16.dlc) | | SalsaNext | w8a16 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 24.488 ms | 1 - 229 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext_w8a16.dlc) | | SalsaNext | w8a16 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 20.052 ms | 1 - 264 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext_w8a16.dlc) | | SalsaNext | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 40.932 ms | 1 - 1 MB | NPU | [SalsaNext.dlc](https://huggingface.co/qualcomm/SalsaNext/blob/main/SalsaNext_w8a16.dlc) | ## Installation Install the package via pip: ```bash pip install qai-hub-models ``` ## Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device Sign-in to [Qualcomm® AI Hub Workbench](https://workbench.aihub.qualcomm.com/) with your Qualcomm® ID. Once signed in navigate to `Account -> Settings -> API Token`. With this API token, you can configure your client to run models on the cloud hosted devices. ```bash qai-hub configure --api_token API_TOKEN ``` Navigate to [docs](https://workbench.aihub.qualcomm.com/docs/) for more information. ## Demo off target The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input. ```bash python -m qai_hub_models.models.salsanext.demo ``` The above demo runs a reference implementation of pre-processing, model inference, and post processing. **NOTE**: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above). ``` %run -m qai_hub_models.models.salsanext.demo ``` ### Run model on a cloud-hosted device In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following: * Performance check on-device on a cloud-hosted device * Downloads compiled assets that can be deployed on-device for Android. * Accuracy check between PyTorch and on-device outputs. ```bash python -m qai_hub_models.models.salsanext.export ``` ## How does this work? This [export script](https://aihub.qualcomm.com/models/salsanext/qai_hub_models/models/SalsaNext/export.py) leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) to optimize, validate, and deploy this model on-device. Lets go through each step below in detail: Step 1: **Compile model for on-device deployment** To compile a PyTorch model for on-device deployment, we first trace the model in memory using the `jit.trace` and then call the `submit_compile_job` API. ```python import torch import qai_hub as hub from qai_hub_models.models.salsanext import Model # Load the model torch_model = Model.from_pretrained() # Device device = hub.Device("Samsung Galaxy S25") # Trace model input_shape = torch_model.get_input_spec() sample_inputs = torch_model.sample_inputs() pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()]) # Compile model on a specific device compile_job = hub.submit_compile_job( model=pt_model, device=device, input_specs=torch_model.get_input_spec(), ) # Get target model to run on-device target_model = compile_job.get_target_model() ``` Step 2: **Performance profiling on cloud-hosted device** After compiling models from step 1. Models can be profiled model on-device using the `target_model`. Note that this scripts runs the model on a device automatically provisioned in the cloud. Once the job is submitted, you can navigate to a provided job URL to view a variety of on-device performance metrics. ```python profile_job = hub.submit_profile_job( model=target_model, device=device, ) ``` Step 3: **Verify on-device accuracy** To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device. ```python input_data = torch_model.sample_inputs() inference_job = hub.submit_inference_job( model=target_model, device=device, inputs=input_data, ) on_device_output = inference_job.download_output_data() ``` With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output. **Note**: This on-device profiling and inference requires access to Qualcomm® AI Hub Workbench. [Sign up for access](https://myaccount.qualcomm.com/signup). ## Deploying compiled model to Android The models can be deployed using multiple runtimes: - TensorFlow Lite (`.tflite` export): [This tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a guide to deploy the .tflite model in an Android application. - QNN (`.so` export ): This [sample app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html) provides instructions on how to use the `.so` shared library in an Android application. ## View on Qualcomm® AI Hub Get more details on SalsaNext's performance across various devices [here](https://aihub.qualcomm.com/models/salsanext). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## 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).