---
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 Qualcomm Devices

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 is based on the implementation of FFNet-78S-LowRes found [here](https://github.com/Qualcomm-AI-research/FFNet).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/ffnet_78s_lowres) 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 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s_lowres/releases/v0.47.0/ffnet_78s_lowres-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s_lowres/releases/v0.47.0/ffnet_78s_lowres-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s_lowres/releases/v0.47.0/ffnet_78s_lowres-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s_lowres/releases/v0.47.0/ffnet_78s_lowres-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s_lowres/releases/v0.47.0/ffnet_78s_lowres-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s_lowres/releases/v0.47.0/ffnet_78s_lowres-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[FFNet-78S-LowRes on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/ffnet_78s_lowres)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/ffnet_78s_lowres) 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 [FFNet-78S-LowRes on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/ffnet_78s_lowres) for usage instructions.

## 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 output classes: 19
- Number of parameters: 26.8M
- Model size (float): 102 MB
- Model size (w8a8): 26.0 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| FFNet-78S-LowRes | ONNX | float | Snapdragon® X Elite | 8.183 ms | 46 - 46 MB | NPU
| FFNet-78S-LowRes | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 5.568 ms | 1 - 210 MB | NPU
| FFNet-78S-LowRes | ONNX | float | Qualcomm® QCS8550 (Proxy) | 7.778 ms | 0 - 51 MB | NPU
| FFNet-78S-LowRes | ONNX | float | Qualcomm® QCS9075 | 12.997 ms | 6 - 15 MB | NPU
| FFNet-78S-LowRes | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.546 ms | 1 - 176 MB | NPU
| FFNet-78S-LowRes | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.822 ms | 7 - 182 MB | NPU
| FFNet-78S-LowRes | ONNX | float | Snapdragon® X2 Elite | 4.202 ms | 47 - 47 MB | NPU
| FFNet-78S-LowRes | ONNX | w8a8 | Snapdragon® X Elite | 3.01 ms | 25 - 25 MB | NPU
| FFNet-78S-LowRes | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 1.977 ms | 0 - 118 MB | NPU
| FFNet-78S-LowRes | ONNX | w8a8 | Qualcomm® QCS6490 | 111.988 ms | 56 - 116 MB | CPU
| FFNet-78S-LowRes | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 2.767 ms | 0 - 29 MB | NPU
| FFNet-78S-LowRes | ONNX | w8a8 | Qualcomm® QCS9075 | 3.32 ms | 1 - 4 MB | NPU
| FFNet-78S-LowRes | ONNX | w8a8 | Qualcomm® QCM6690 | 118.793 ms | 59 - 69 MB | CPU
| FFNet-78S-LowRes | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 1.503 ms | 0 - 65 MB | NPU
| FFNet-78S-LowRes | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 119.739 ms | 57 - 68 MB | CPU
| FFNet-78S-LowRes | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 1.27 ms | 0 - 61 MB | NPU
| FFNet-78S-LowRes | ONNX | w8a8 | Snapdragon® X2 Elite | 1.321 ms | 25 - 25 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Snapdragon® X Elite | 15.314 ms | 6 - 6 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 9.772 ms | 6 - 84 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 51.189 ms | 1 - 44 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 14.577 ms | 6 - 8 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Qualcomm® SA8775P | 19.155 ms | 1 - 44 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Qualcomm® QCS9075 | 18.226 ms | 6 - 14 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 26.56 ms | 5 - 73 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Qualcomm® SA7255P | 51.189 ms | 1 - 44 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Qualcomm® SA8295P | 21.187 ms | 0 - 40 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 8.122 ms | 0 - 48 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 6.635 ms | 6 - 54 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | float | Snapdragon® X2 Elite | 7.609 ms | 6 - 6 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Snapdragon® X Elite | 5.046 ms | 2 - 2 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 3.311 ms | 2 - 103 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 14.872 ms | 2 - 5 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 10.513 ms | 2 - 56 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 4.671 ms | 2 - 3 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Qualcomm® SA8775P | 5.175 ms | 2 - 57 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 6.848 ms | 3 - 7 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 36.605 ms | 2 - 191 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 7.223 ms | 2 - 100 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Qualcomm® SA7255P | 10.513 ms | 2 - 56 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Qualcomm® SA8295P | 6.529 ms | 1 - 53 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 2.234 ms | 2 - 57 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 7.335 ms | 2 - 184 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 1.852 ms | 2 - 56 MB | NPU
| FFNet-78S-LowRes | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 2.173 ms | 2 - 2 MB | NPU
| FFNet-78S-LowRes | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 9.897 ms | 1 - 161 MB | NPU
| FFNet-78S-LowRes | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 51.039 ms | 1 - 78 MB | NPU
| FFNet-78S-LowRes | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 14.736 ms | 0 - 3 MB | NPU
| FFNet-78S-LowRes | TFLITE | float | Qualcomm® SA8775P | 19.243 ms | 1 - 79 MB | NPU
| FFNet-78S-LowRes | TFLITE | float | Qualcomm® QCS9075 | 18.364 ms | 0 - 60 MB | NPU
| FFNet-78S-LowRes | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 26.249 ms | 1 - 150 MB | NPU
| FFNet-78S-LowRes | TFLITE | float | Qualcomm® SA7255P | 51.039 ms | 1 - 78 MB | NPU
| FFNet-78S-LowRes | TFLITE | float | Qualcomm® SA8295P | 20.969 ms | 1 - 69 MB | NPU
| FFNet-78S-LowRes | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 8.172 ms | 0 - 78 MB | NPU
| FFNet-78S-LowRes | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 6.638 ms | 1 - 81 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 1.978 ms | 0 - 106 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Qualcomm® QCS6490 | 9.548 ms | 0 - 30 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 7.299 ms | 0 - 52 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 2.787 ms | 0 - 2 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Qualcomm® SA8775P | 3.251 ms | 0 - 54 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Qualcomm® QCS9075 | 3.267 ms | 0 - 29 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Qualcomm® QCM6690 | 27.538 ms | 0 - 184 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 3.506 ms | 0 - 101 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Qualcomm® SA7255P | 7.299 ms | 0 - 52 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Qualcomm® SA8295P | 4.454 ms | 0 - 50 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 1.489 ms | 0 - 54 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 4.2 ms | 0 - 181 MB | NPU
| FFNet-78S-LowRes | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 1.299 ms | 0 - 53 MB | NPU

## 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).

## 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).