Hugging Face's logo

qualcomm
/
FFNet-78S

Image Segmentation
PyTorch
android
Model card Files
xet
 Community
FFNet-78S
File size: 10,185 Bytes
aa00d2e
 
ccb4707
aa00d2e
 
cbf6b66
aa00d2e
 
 
eb9fa30
aa00d2e
169c686
f5f2058
aa00d2e
 
169c686
25cd5b7
169c686
 
 
 
 
 
 
 
 
 
 
 
4e034ab
 
 
 
 
 
169c686
 
 
 
 
 
25cd5b7
169c686
 
 
 
 
 
25cd5b7
169c686
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4e034ab
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2c76d38
aa00d2e
db5fb2d
 
2c76d38
aa00d2e
 
 
 
 
eb7d1e0
aa00d2e
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
---
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/main/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 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s/releases/v0.51.0/ffnet_78s-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/ffnet_78s/releases/v0.51.0/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/v0.51.0/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/v0.51.0/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/v0.51.0/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/v0.51.0/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/main/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/main/src/qai_hub_models/models/ffnet_78s) for usage instructions.

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

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

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| FFNet-78S | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 17.173 ms | 29 - 258 MB | NPU
| FFNet-78S | ONNX | float | Snapdragon® X2 Elite | 17.872 ms | 30 - 30 MB | NPU
| FFNet-78S | ONNX | float | Snapdragon® X Elite | 37.856 ms | 30 - 30 MB | NPU
| FFNet-78S | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 27.043 ms | 0 - 301 MB | NPU
| FFNet-78S | ONNX | float | Qualcomm® QCS8550 (Proxy) | 38.338 ms | 24 - 27 MB | NPU
| FFNet-78S | ONNX | float | Qualcomm® QCS9075 | 59.331 ms | 24 - 51 MB | NPU
| FFNet-78S | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 20.224 ms | 6 - 209 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 7.282 ms | 0 - 219 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® X2 Elite | 7.925 ms | 22 - 22 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® X Elite | 14.884 ms | 21 - 21 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 11.938 ms | 7 - 295 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® QCS6490 | 507.691 ms | 162 - 230 MB | CPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 15.421 ms | 0 - 102 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® QCS9075 | 14.482 ms | 6 - 9 MB | NPU
| FFNet-78S | ONNX | w8a8 | Qualcomm® QCM6690 | 533.377 ms | 139 - 148 MB | CPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 8.78 ms | 1 - 209 MB | NPU
| FFNet-78S | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 532.845 ms | 148 - 158 MB | CPU
| FFNet-78S | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.463 ms | 24 - 273 MB | NPU
| FFNet-78S | QNN_DLC | float | Snapdragon® X2 Elite | 17.953 ms | 24 - 24 MB | NPU
| FFNet-78S | QNN_DLC | float | Snapdragon® X Elite | 44.25 ms | 24 - 24 MB | NPU
| FFNet-78S | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 29.407 ms | 19 - 315 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 187.95 ms | 24 - 233 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 42.6 ms | 24 - 26 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® SA8775P | 60.845 ms | 24 - 234 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® QCS9075 | 73.42 ms | 24 - 52 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 86.396 ms | 4 - 285 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® SA7255P | 187.95 ms | 24 - 233 MB | NPU
| FFNet-78S | QNN_DLC | float | Qualcomm® SA8295P | 66.496 ms | 24 - 228 MB | NPU
| FFNet-78S | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 21.845 ms | 24 - 254 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 5.884 ms | 6 - 250 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 7.009 ms | 6 - 6 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® X Elite | 17.604 ms | 6 - 6 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 11.609 ms | 6 - 291 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 72.281 ms | 8 - 16 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 39.065 ms | 6 - 225 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 16.781 ms | 6 - 8 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® SA8775P | 17.196 ms | 6 - 226 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 20.011 ms | 6 - 14 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 165.642 ms | 6 - 256 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 23.657 ms | 6 - 293 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® SA7255P | 39.065 ms | 6 - 225 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Qualcomm® SA8295P | 23.328 ms | 6 - 227 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 8.06 ms | 6 - 238 MB | NPU
| FFNet-78S | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 21.825 ms | 6 - 242 MB | NPU
| FFNet-78S | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.483 ms | 2 - 281 MB | NPU
| FFNet-78S | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 29.449 ms | 2 - 378 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 187.958 ms | 3 - 245 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 42.887 ms | 2 - 5 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® SA8775P | 61.023 ms | 2 - 245 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® QCS9075 | 73.561 ms | 0 - 82 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 88.15 ms | 3 - 362 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® SA7255P | 187.958 ms | 3 - 245 MB | NPU
| FFNet-78S | TFLITE | float | Qualcomm® SA8295P | 66.475 ms | 2 - 240 MB | NPU
| FFNet-78S | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 21.775 ms | 2 - 264 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 3.578 ms | 0 - 242 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 6.465 ms | 1 - 290 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® QCS6490 | 57.526 ms | 1 - 36 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 26.245 ms | 1 - 218 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 9.068 ms | 1 - 3 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® SA8775P | 9.723 ms | 1 - 219 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® QCS9075 | 11.161 ms | 1 - 36 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® QCM6690 | 139.037 ms | 1 - 248 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 18.45 ms | 1 - 291 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® SA7255P | 26.245 ms | 1 - 218 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Qualcomm® SA8295P | 14.886 ms | 1 - 221 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 5.04 ms | 1 - 231 MB | NPU
| FFNet-78S | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 14.574 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).