README.md

---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: image-segmentation

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s/web-assets/model_demo.png)

# FFNet-78S: Optimized for Qualcomm Devices

FFNet-78S 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 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/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/ffnet_78s) 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 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s/releases/v0.60.0/ffnet_78s-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s/releases/v0.60.0/ffnet_78s-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s/releases/v0.60.0/ffnet_78s-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s/releases/v0.60.0/ffnet_78s-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s/releases/v0.60.0/ffnet_78s-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s/releases/v0.60.0/ffnet_78s-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[FFNet-78S on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/ffnet_78s)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/ffnet_78s) 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 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/ffnet_78s) for usage instructions.

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

**Model Stats:**
- Input resolution: 2048x1024
- Model checkpoint: ffnet78S_dBBB_cityscapes_state_dict_quarts
- Model size (float): 105 MB
- Model size (w8a8): 26.7 MB
- Number of output classes: 19
- Number of parameters: 27.5M

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| FFNet-78S | ONNX | float | Snapdragon® X2 Elite | 18.406 ms | 24 - 24 MB | NPU
| FFNet-78S | ONNX | float | Snapdragon® X Elite | 37.496 ms | 31 - 31 MB | NPU
| FFNet-78S | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 27.346 ms | 0 - 316 MB | NPU
| FFNet-78S | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 60.18 ms | 13 - 296 MB | NPU
| FFNet-78S | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 58.076 ms | 24 - 51 MB | NPU
| FFNet-78S | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 38.642 ms | 24 - 54 MB | NPU
| FFNet-78S | ONNX | float | Qualcomm® QCS8450 | 60.18 ms | 13 - 296 MB | NPU
| FFNet-78S | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 61.637 ms | 24 - 51 MB | NPU
| FFNet-78S | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 37.496 ms | 31 - 31 MB | NPU
| FFNet-78S | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 21.651 ms | 8 - 207 MB | NPU
| FFNet-78S | ONNX | float | Snapdragon® 8 Elite Mobile | 21.651 ms | 8 - 207 MB | NPU
| FFNet-78S | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 17.798 ms | 6 - 229 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® X2 Elite | 9.31 ms | 7 - 7 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® X Elite | 13.6 ms | 22 - 22 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 8.551 ms | 7 - 301 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 17.807 ms | 8 - 301 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 57.086 ms | 6 - 9 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 12.802 ms | 6 - 10 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 12.838 ms | 0 - 24 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® QCS8450 | 17.807 ms | 8 - 301 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 13.501 ms | 6 - 9 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 13.6 ms | 22 - 22 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 150.009 ms | 6 - 237 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 17.779 ms | 7 - 257 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 8.741 ms | 1 - 216 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 8.741 ms | 1 - 216 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 7.413 ms | 2 - 219 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 17.779 ms | 7 - 257 MB | NPU
| FFNet-78S | QNN_DLC | float | Snapdragon® X2 Elite | 17.97 ms | 24 - 24 MB | NPU
| FFNet-78S | QNN_DLC | float | Snapdragon® X Elite | 44.558 ms | 24 - 24 MB | NPU
| FFNet-78S | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 29.695 ms | 19 - 315 MB | NPU
| FFNet-78S | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 85.9 ms | 5 - 286 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 62.48 ms | 24 - 53 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 187.969 ms | 24 - 234 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 42.756 ms | 24 - 35 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® SA8775P | 60.877 ms | 24 - 234 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® SA8650P | 60.877 ms | 24 - 234 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® SA8255P | 60.877 ms | 24 - 234 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® QCS8450 | 85.9 ms | 5 - 286 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 61.367 ms | 24 - 52 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 44.558 ms | 24 - 24 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 21.74 ms | 12 - 238 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® SA7255P | 187.969 ms | 24 - 234 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® SA8295P | 66.507 ms | 24 - 228 MB | NPU
| FFNet-78S | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 21.74 ms | 12 - 238 MB | NPU
| FFNet-78S | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.635 ms | 14 - 267 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 6.977 ms | 6 - 6 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® X Elite | 17.785 ms | 6 - 6 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 11.613 ms | 6 - 290 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 23.625 ms | 6 - 292 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 63.946 ms | 3 - 12 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 39.091 ms | 6 - 226 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 16.776 ms | 6 - 8 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® SA8775P | 17.186 ms | 6 - 226 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® SA8650P | 17.186 ms | 6 - 226 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® SA8255P | 17.186 ms | 6 - 226 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 23.625 ms | 6 - 292 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 17.446 ms | 8 - 16 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 17.785 ms | 6 - 6 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 166.357 ms | 6 - 256 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 22.32 ms | 6 - 243 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 8.055 ms | 0 - 231 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® SA7255P | 39.091 ms | 6 - 226 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® SA8295P | 23.333 ms | 6 - 227 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 8.055 ms | 0 - 231 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 5.967 ms | 6 - 251 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 22.32 ms | 6 - 243 MB | NPU
| FFNet-78S | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 29.431 ms | 1 - 377 MB | NPU
| FFNet-78S | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 86.118 ms | 3 - 363 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 62.554 ms | 0 - 83 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 188.047 ms | 3 - 245 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 42.889 ms | 3 - 5 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® SA8775P | 60.992 ms | 3 - 245 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® SA8650P | 60.992 ms | 3 - 245 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® SA8255P | 60.992 ms | 3 - 245 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® QCS8450 | 86.118 ms | 3 - 363 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 67.349 ms | 0 - 82 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 21.727 ms | 2 - 258 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® SA7255P | 188.047 ms | 3 - 245 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® SA8295P | 66.385 ms | 3 - 239 MB | NPU
| FFNet-78S | TFLITE | float | Snapdragon® 8 Elite Mobile | 21.727 ms | 2 - 258 MB | NPU
| FFNet-78S | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.675 ms | 0 - 281 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 6.43 ms | 1 - 287 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 18.283 ms | 1 - 291 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 49.896 ms | 1 - 36 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 9.77 ms | 1 - 37 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 26.206 ms | 1 - 218 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 9.063 ms | 1 - 3 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® SA8775P | 9.739 ms | 1 - 219 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® SA8650P | 9.739 ms | 1 - 219 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® SA8255P | 9.739 ms | 1 - 219 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® QCS8450 | 18.283 ms | 1 - 291 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 9.462 ms | 1 - 36 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 139.928 ms | 1 - 246 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 14.724 ms | 1 - 234 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 5.036 ms | 1 - 231 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® SA7255P | 26.206 ms | 1 - 218 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 5.036 ms | 1 - 231 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 3.654 ms | 1 - 243 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 14.724 ms | 1 - 234 MB | NPU

## License
* The license for the original implementation of FFNet-78S 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).