--- library_name: pytorch license: bsd-3-clause pipeline_tag: image-segmentation tags: - quantized - real_time - android --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s_quantized/web-assets/model_demo.png) # FFNet-78S-Quantized: Optimized for Mobile Deployment ## Semantic segmentation for automotive street scenes FFNet-78S-Quantized is a "fuss-free network" that segments street scene images with per-pixel classes like road, sidewalk, and pedestrian. Trained on the Cityscapes dataset. This model is an implementation of FFNet-78S-Quantized found [here](https://github.com/Qualcomm-AI-research/FFNet). This repository provides scripts to run FFNet-78S-Quantized on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/ffnet_78s_quantized). ### Model Details - **Model Type:** Semantic segmentation - **Model Stats:** - Model checkpoint: ffnet78S_dBBB_cityscapes_state_dict_quarts - Input resolution: 2048x1024 - Number of parameters: 27.5M - Model size: 26.7 MB - Number of output classes: 19 | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | FFNet-78S-Quantized | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 5.84 ms | 1 - 3 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 11.928 ms | 0 - 367 MB | INT8 | NPU | [FFNet-78S-Quantized.onnx](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.onnx) | | FFNet-78S-Quantized | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 4.034 ms | 1 - 88 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 8.243 ms | 7 - 157 MB | INT8 | NPU | [FFNet-78S-Quantized.onnx](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.onnx) | | FFNet-78S-Quantized | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 4.055 ms | 0 - 38 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 7.065 ms | 2 - 71 MB | INT8 | NPU | [FFNet-78S-Quantized.onnx](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.onnx) | | FFNet-78S-Quantized | RB3 Gen 2 (Proxy) | QCS6490 Proxy | TFLITE | 35.537 ms | 0 - 47 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | RB5 (Proxy) | QCS8250 Proxy | TFLITE | 219.116 ms | 1 - 9 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 5.684 ms | 1 - 2 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 5.742 ms | 0 - 2 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 5.818 ms | 1 - 3 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 5.788 ms | 1 - 3 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | SA8295P ADP | SA8295P | TFLITE | 11.213 ms | 1 - 38 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 7.082 ms | 1 - 90 MB | INT8 | NPU | [FFNet-78S-Quantized.tflite](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.tflite) | | FFNet-78S-Quantized | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 12.451 ms | 23 - 23 MB | INT8 | NPU | [FFNet-78S-Quantized.onnx](https://huggingface.co/qualcomm/FFNet-78S-Quantized/blob/main/FFNet-78S-Quantized.onnx) | ## Installation This model can be installed as a Python package via pip. ```bash pip install "qai-hub-models[ffnet_78s_quantized]" ``` ## Configure Qualcomm® AI Hub to run this model on a cloud-hosted device Sign-in to [Qualcomm® AI Hub](https://app.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://app.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.ffnet_78s_quantized.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.ffnet_78s_quantized.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.ffnet_78s_quantized.export ``` ``` Profiling Results ------------------------------------------------------------ FFNet-78S-Quantized Device : Samsung Galaxy S23 (13) Runtime : TFLITE Estimated inference time (ms) : 5.8 Estimated peak memory usage (MB): [1, 3] Total # Ops : 156 Compute Unit(s) : NPU (156 ops) ``` ## 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 FFNet-78S-Quantized's performance across various devices [here](https://aihub.qualcomm.com/models/ffnet_78s_quantized). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of FFNet-78S-Quantized can be found [here](https://github.com/Qualcomm-AI-research/FFNet/blob/master/LICENSE). * The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf) ## References * [Simple and Efficient Architectures for Semantic Segmentation](https://arxiv.org/abs/2206.08236) * [Source Model Implementation](https://github.com/Qualcomm-AI-research/FFNet) ## 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).