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_54s/web-assets/model_demo.png)

# FFNet-54S: Optimized for Qualcomm Devices

FFNet-54S 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-54S 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_54s) 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_54s/releases/v0.60.0/ffnet_54s-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_54s/releases/v0.60.0/ffnet_54s-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_54s/releases/v0.60.0/ffnet_54s-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_54s/releases/v0.60.0/ffnet_54s-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_54s/releases/v0.60.0/ffnet_54s-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_54s/releases/v0.60.0/ffnet_54s-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[FFNet-54S on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/ffnet_54s)**.


### 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_54s) 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-54S on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/ffnet_54s) for usage instructions.

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

**Model Stats:**
- Input resolution: 2048x1024
- Model checkpoint: ffnet54S_dBBB_cityscapes_state_dict_quarts
- Model size (float): 68.8 MB
- Model size (w8a8): 17.5 MB
- Number of output classes: 19
- Number of parameters: 18.0M

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| FFNet-54S | ONNX | float | Snapdragon® X2 Elite | 15.532 ms | 24 - 24 MB | NPU
| FFNet-54S | ONNX | float | Snapdragon® X Elite | 33.938 ms | 24 - 24 MB | NPU
| FFNet-54S | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 24.621 ms | 5 - 306 MB | NPU
| FFNet-54S | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 53.206 ms | 0 - 270 MB | NPU
| FFNet-54S | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 49.477 ms | 24 - 51 MB | NPU
| FFNet-54S | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 34.135 ms | 24 - 39 MB | NPU
| FFNet-54S | ONNX | float | Qualcomm® QCS8450 | 53.206 ms | 0 - 270 MB | NPU
| FFNet-54S | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 50.061 ms | 24 - 51 MB | NPU
| FFNet-54S | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 33.938 ms | 24 - 24 MB | NPU
| FFNet-54S | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 19.526 ms | 6 - 197 MB | NPU
| FFNet-54S | ONNX | float | Snapdragon® 8 Elite Mobile | 19.526 ms | 6 - 197 MB | NPU
| FFNet-54S | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 14.61 ms | 5 - 220 MB | NPU
| FFNet-54S | ONNX | w8a8 | Snapdragon® X2 Elite | 7.399 ms | 7 - 7 MB | NPU
| FFNet-54S | ONNX | w8a8 | Snapdragon® X Elite | 11.888 ms | 12 - 12 MB | NPU
| FFNet-54S | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 7.359 ms | 7 - 272 MB | NPU
| FFNet-54S | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 16.814 ms | 7 - 270 MB | NPU
| FFNet-54S | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 52.069 ms | 6 - 9 MB | NPU
| FFNet-54S | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 10.835 ms | 6 - 10 MB | NPU
| FFNet-54S | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 11.311 ms | 0 - 195 MB | NPU
| FFNet-54S | ONNX | w8a8 | Qualcomm® QCS8450 | 16.814 ms | 7 - 270 MB | NPU
| FFNet-54S | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 11.728 ms | 6 - 9 MB | NPU
| FFNet-54S | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 11.888 ms | 12 - 12 MB | NPU
| FFNet-54S | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 123.529 ms | 6 - 255 MB | NPU
| FFNet-54S | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 14.962 ms | 7 - 223 MB | NPU
| FFNet-54S | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 11.758 ms | 1 - 201 MB | NPU
| FFNet-54S | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 11.758 ms | 1 - 201 MB | NPU
| FFNet-54S | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 6.994 ms | 2 - 205 MB | NPU
| FFNet-54S | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 14.962 ms | 7 - 223 MB | NPU
| FFNet-54S | QNN_DLC | float | Snapdragon® X2 Elite | 15.885 ms | 24 - 24 MB | NPU
| FFNet-54S | QNN_DLC | float | Snapdragon® X Elite | 39.885 ms | 24 - 24 MB | NPU
| FFNet-54S | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 26.687 ms | 24 - 302 MB | NPU
| FFNet-54S | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 77.114 ms | 24 - 297 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 53.776 ms | 24 - 53 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 155.454 ms | 24 - 225 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 38.539 ms | 24 - 26 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® SA8775P | 53.667 ms | 24 - 225 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® SA8650P | 53.667 ms | 24 - 225 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® SA8255P | 53.667 ms | 24 - 225 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® QCS8450 | 77.114 ms | 24 - 297 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 54.369 ms | 24 - 52 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 39.885 ms | 24 - 24 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 19.649 ms | 23 - 245 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® SA7255P | 155.454 ms | 24 - 225 MB | NPU
| FFNet-54S | QNN_DLC | float | Qualcomm® SA8295P | 59.263 ms | 24 - 222 MB | NPU
| FFNet-54S | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 19.649 ms | 23 - 245 MB | NPU
| FFNet-54S | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 13.814 ms | 18 - 262 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 6.441 ms | 6 - 6 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Snapdragon® X Elite | 16.754 ms | 6 - 6 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 10.94 ms | 6 - 264 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 22.209 ms | 6 - 267 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 58.685 ms | 4 - 12 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 15.542 ms | 3 - 12 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 35.57 ms | 6 - 212 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 15.642 ms | 6 - 32 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® SA8775P | 16.281 ms | 6 - 212 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® SA8650P | 16.281 ms | 6 - 212 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® SA8255P | 16.281 ms | 6 - 212 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 22.209 ms | 6 - 267 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 16.391 ms | 6 - 14 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 16.754 ms | 6 - 6 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 140.158 ms | 6 - 244 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 20.171 ms | 6 - 229 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 7.465 ms | 6 - 224 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® SA7255P | 35.57 ms | 6 - 212 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Qualcomm® SA8295P | 21.299 ms | 6 - 215 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 7.465 ms | 6 - 224 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 5.562 ms | 6 - 238 MB | NPU
| FFNet-54S | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 20.171 ms | 6 - 229 MB | NPU
| FFNet-54S | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 26.835 ms | 0 - 330 MB | NPU
| FFNet-54S | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 77.039 ms | 3 - 324 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 53.859 ms | 0 - 65 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 155.473 ms | 3 - 224 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 39.078 ms | 2 - 5 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® SA8775P | 53.722 ms | 3 - 224 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® SA8650P | 53.722 ms | 3 - 224 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® SA8255P | 53.722 ms | 3 - 224 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® QCS8450 | 77.039 ms | 3 - 324 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 60.672 ms | 0 - 64 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 19.806 ms | 2 - 246 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® SA7255P | 155.473 ms | 3 - 224 MB | NPU
| FFNet-54S | TFLITE | float | Qualcomm® SA8295P | 59.241 ms | 0 - 217 MB | NPU
| FFNet-54S | TFLITE | float | Snapdragon® 8 Elite Mobile | 19.806 ms | 2 - 246 MB | NPU
| FFNet-54S | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 13.67 ms | 2 - 264 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 5.928 ms | 1 - 261 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 13.292 ms | 1 - 263 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 44.642 ms | 0 - 27 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 8.921 ms | 1 - 28 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 22.867 ms | 1 - 203 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 8.261 ms | 1 - 3 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® SA8775P | 8.828 ms | 1 - 205 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® SA8650P | 8.828 ms | 1 - 205 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® SA8255P | 8.828 ms | 1 - 205 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® QCS8450 | 13.292 ms | 1 - 263 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 8.602 ms | 1 - 27 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 113.899 ms | 1 - 236 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 12.586 ms | 1 - 220 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 4.459 ms | 1 - 222 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® SA7255P | 22.867 ms | 1 - 203 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Qualcomm® SA8295P | 13.18 ms | 0 - 206 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 4.459 ms | 1 - 222 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 3.236 ms | 1 - 233 MB | NPU
| FFNet-54S | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 12.586 ms | 1 - 220 MB | NPU

## License
* The license for the original implementation of FFNet-54S 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).