--- library_name: pytorch license: other tags: - real_time - android pipeline_tag: image-segmentation --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s_lowres/web-assets/model_demo.png) # FFNet-78S-LowRes: Optimized for Mobile Deployment ## Semantic segmentation for automotive street scenes FFNet-78S-LowRes 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-LowRes found [here](https://github.com/Qualcomm-AI-research/FFNet). This repository provides scripts to run FFNet-78S-LowRes on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/ffnet_78s_lowres). ### Model Details - **Model Type:** Model_use_case.semantic_segmentation - **Model Stats:** - Model checkpoint: ffnet78S_BCC_cityscapes_state_dict_quarts_pre_down - Input resolution: 1024x512 - Number of parameters: 26.8M - Model size: 102 MB - Number of output classes: 19 | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | FFNet-78S-LowRes | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 50.335 ms | 1 - 61 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 48.043 ms | 6 - 35 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 17.267 ms | 1 - 115 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 25.641 ms | 3 - 42 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 13.161 ms | 1 - 13 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 12.325 ms | 6 - 15 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 17.79 ms | 1 - 61 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 16.859 ms | 0 - 29 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 50.335 ms | 1 - 61 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 48.043 ms | 6 - 35 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 13.168 ms | 1 - 15 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 12.623 ms | 6 - 15 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 19.839 ms | 0 - 57 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 18.68 ms | 3 - 30 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 13.163 ms | 0 - 32 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 12.425 ms | 6 - 16 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 17.79 ms | 1 - 61 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 16.859 ms | 0 - 29 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 13.208 ms | 1 - 19 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 12.4 ms | 6 - 15 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 9.271 ms | 2 - 151 MB | NPU | [FFNet-78S-LowRes.onnx](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.onnx) | | FFNet-78S-LowRes | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 9.091 ms | 1 - 122 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 8.481 ms | 4 - 45 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 7.284 ms | 8 - 58 MB | NPU | [FFNet-78S-LowRes.onnx](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.onnx) | | FFNet-78S-LowRes | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 9.194 ms | 1 - 64 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.tflite) | | FFNet-78S-LowRes | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 8.57 ms | 1 - 35 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 5.8 ms | 6 - 43 MB | NPU | [FFNet-78S-LowRes.onnx](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.onnx) | | FFNet-78S-LowRes | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 18.259 ms | 98 - 98 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.dlc) | | FFNet-78S-LowRes | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 9.661 ms | 47 - 47 MB | NPU | [FFNet-78S-LowRes.onnx](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes.onnx) | | FFNet-78S-LowRes | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 7.766 ms | 0 - 38 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 10.549 ms | 1 - 41 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 3.423 ms | 0 - 73 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 6.612 ms | 2 - 72 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 3.003 ms | 0 - 197 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 4.638 ms | 1 - 16 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 3.503 ms | 0 - 39 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 5.12 ms | 2 - 41 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 10.581 ms | 0 - 62 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN_DLC | 17.755 ms | 2 - 60 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 65.812 ms | 12 - 25 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 7.766 ms | 0 - 38 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 10.549 ms | 1 - 41 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 3.01 ms | 0 - 199 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 4.659 ms | 1 - 153 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 4.68 ms | 0 - 42 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 6.506 ms | 2 - 45 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 3.046 ms | 0 - 198 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 4.651 ms | 2 - 165 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 3.503 ms | 0 - 39 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 5.12 ms | 2 - 41 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 3.046 ms | 0 - 195 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 4.646 ms | 2 - 155 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 4.88 ms | 0 - 73 MB | NPU | [FFNet-78S-LowRes.onnx](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.onnx) | | FFNet-78S-LowRes | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 2.193 ms | 0 - 71 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 3.317 ms | 2 - 71 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 3.441 ms | 6 - 115 MB | NPU | [FFNet-78S-LowRes.onnx](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.onnx) | | FFNet-78S-LowRes | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 2.132 ms | 0 - 39 MB | NPU | [FFNet-78S-LowRes.tflite](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.tflite) | | FFNet-78S-LowRes | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 2.669 ms | 2 - 46 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 3.443 ms | 1 - 112 MB | NPU | [FFNet-78S-LowRes.onnx](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.onnx) | | FFNet-78S-LowRes | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 6.364 ms | 188 - 188 MB | NPU | [FFNet-78S-LowRes.dlc](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.dlc) | | FFNet-78S-LowRes | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 5.256 ms | 24 - 24 MB | NPU | [FFNet-78S-LowRes.onnx](https://huggingface.co/qualcomm/FFNet-78S-LowRes/blob/main/FFNet-78S-LowRes_w8a8.onnx) | ## Installation Install the package via pip: ```bash pip install "qai-hub-models[ffnet-78s-lowres]" ``` ## 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_lowres.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_lowres.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_lowres.export ``` ``` Profiling Results ------------------------------------------------------------ FFNet-78S-LowRes Device : cs_8275 (ANDROID 14) Runtime : TFLITE Estimated inference time (ms) : 50.3 Estimated peak memory usage (MB): [1, 61] Total # Ops : 151 Compute Unit(s) : npu (151 ops) gpu (0 ops) cpu (0 ops) ``` ## How does this work? This [export script](https://aihub.qualcomm.com/models/ffnet_78s_lowres/qai_hub_models/models/FFNet-78S-LowRes/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.ffnet_78s_lowres import Model # Load the model torch_model = Model.from_pretrained() # Device device = hub.Device("Samsung Galaxy S24") # 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. [Sign up for access](https://myaccount.qualcomm.com/signup). ## Run demo on a cloud-hosted device You can also run the demo on-device. ```bash python -m qai_hub_models.models.ffnet_78s_lowres.demo --eval-mode on-device ``` **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_lowres.demo -- --eval-mode on-device ``` ## 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-LowRes's performance across various devices [here](https://aihub.qualcomm.com/models/ffnet_78s_lowres). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of FFNet-78S-LowRes 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).