Hugging Face's logo

qualcomm
/
PidNet

Image Segmentation
PyTorch
real_time
android
Model card Files
xet
 Community
PidNet
File size: 11,773 Bytes
2a7fd96
 
a3151a5
2a7fd96
 
 
 
 
 
 
 
 
36ff84e
2a7fd96
 
 
36ff84e
d22b436
36ff84e
 
 
 
 
 
 
 
 
 
 
 
d22b436
 
 
 
 
 
36ff84e
 
 
 
 
 
d22b436
36ff84e
 
 
 
 
 
d22b436
36ff84e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d22b436
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2a7fd96
 
 
 
 
 
 
 
 
 
 
 
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
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
---
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/pidnet/web-assets/model_demo.png)

# PidNet: Optimized for Qualcomm Devices

PIDNet (Proportional-Integral-Derivative Network) is a real-time semantic segmentation model based on  PID controllers

This is based on the implementation of PidNet found [here](https://github.com/XuJiacong/PIDNet).
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/v0.59.0/src/qai_hub_models/models/pidnet) 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/pidnet/releases/v0.59.0/pidnet-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/pidnet/releases/v0.59.0/pidnet-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/pidnet/releases/v0.59.0/pidnet-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/pidnet/releases/v0.59.0/pidnet-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/pidnet/releases/v0.59.0/pidnet-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pidnet/releases/v0.59.0/pidnet-tflite-w8a8.zip)

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


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/pidnet) 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 [PidNet on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/pidnet) for usage instructions.

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

**Model Stats:**
- Model checkpoint: PIDNet_S_Cityscapes_val.pt
- Inference latency: RealTime
- Input resolution: 1024x2048
- Number of output classes: 19
- Number of parameters: 8.06M
- Model size (float): 29.1 MB
- Model size (w8a8): 8.02 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| PidNet | ONNX | float | Snapdragon® X2 Elite | 13.744 ms | 24 - 24 MB | NPU
| PidNet | ONNX | float | Snapdragon® X Elite | 32.283 ms | 24 - 24 MB | NPU
| PidNet | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 23.087 ms | 30 - 331 MB | NPU
| PidNet | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 55.313 ms | 26 - 332 MB | NPU
| PidNet | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 32.647 ms | 24 - 46 MB | NPU
| PidNet | ONNX | float | Qualcomm® QCS8450 | 55.313 ms | 26 - 332 MB | NPU
| PidNet | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 49.376 ms | 24 - 51 MB | NPU
| PidNet | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 32.283 ms | 24 - 24 MB | NPU
| PidNet | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 17.624 ms | 8 - 220 MB | NPU
| PidNet | ONNX | float | Snapdragon® 8 Elite Mobile | 17.624 ms | 8 - 220 MB | NPU
| PidNet | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 13.254 ms | 7 - 257 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® X2 Elite | 25.929 ms | 7 - 7 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® X Elite | 55.07 ms | 7 - 7 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 39.151 ms | 7 - 271 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 57.854 ms | 8 - 271 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 158.684 ms | 6 - 9 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 52.548 ms | 6 - 12 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® QCS8450 | 57.854 ms | 8 - 271 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 53.854 ms | 6 - 9 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 55.07 ms | 7 - 7 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 45.21 ms | 2 - 218 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 45.21 ms | 2 - 218 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 25.835 ms | 2 - 237 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® X2 Elite | 13.648 ms | 24 - 24 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® X Elite | 38.863 ms | 24 - 24 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 25.44 ms | 15 - 321 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 75.769 ms | 4 - 310 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 37.314 ms | 24 - 306 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® SA8775P | 46.572 ms | 24 - 243 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® SA8650P | 46.572 ms | 24 - 243 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® SA8255P | 46.572 ms | 24 - 243 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® QCS8450 | 75.769 ms | 4 - 310 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 47.54 ms | 26 - 54 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 38.863 ms | 24 - 24 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 18.3 ms | 23 - 268 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® SA8295P | 51.999 ms | 24 - 254 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 18.3 ms | 23 - 268 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 12.133 ms | 16 - 288 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 26.231 ms | 6 - 6 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® X Elite | 60.559 ms | 6 - 6 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 43.014 ms | 6 - 268 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 63.841 ms | 6 - 268 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8275 | 111.372 ms | 6 - 218 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 57.836 ms | 6 - 9 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® SA8775P | 58.506 ms | 6 - 219 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® SA8650P | 58.506 ms | 6 - 219 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® SA8255P | 58.506 ms | 6 - 219 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 63.841 ms | 6 - 268 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 66.963 ms | 4 - 13 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 60.559 ms | 6 - 6 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 43.862 ms | 6 - 243 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® SA7255P | 111.372 ms | 6 - 218 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® SA8295P | 66.436 ms | 6 - 223 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 43.862 ms | 6 - 243 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 24.603 ms | 6 - 272 MB | NPU
| PidNet | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 24.998 ms | 1 - 320 MB | NPU
| PidNet | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 74.636 ms | 0 - 325 MB | NPU
| PidNet | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 116.584 ms | 3 - 231 MB | NPU
| PidNet | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 36.628 ms | 2 - 25 MB | NPU
| PidNet | TFLITE | float | Qualcomm® SA8775P | 45.335 ms | 0 - 229 MB | NPU
| PidNet | TFLITE | float | Qualcomm® SA8650P | 45.335 ms | 0 - 229 MB | NPU
| PidNet | TFLITE | float | Qualcomm® SA8255P | 45.335 ms | 0 - 229 MB | NPU
| PidNet | TFLITE | float | Qualcomm® QCS8450 | 74.636 ms | 0 - 325 MB | NPU
| PidNet | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 46.698 ms | 0 - 45 MB | NPU
| PidNet | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 18.786 ms | 2 - 252 MB | NPU
| PidNet | TFLITE | float | Qualcomm® SA7255P | 116.584 ms | 3 - 231 MB | NPU
| PidNet | TFLITE | float | Qualcomm® SA8295P | 51.012 ms | 3 - 245 MB | NPU
| PidNet | TFLITE | float | Snapdragon® 8 Elite Mobile | 18.786 ms | 2 - 252 MB | NPU
| PidNet | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 12.867 ms | 2 - 273 MB | NPU
| PidNet | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 38.042 ms | 0 - 265 MB | NPU
| PidNet | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 58.248 ms | 1 - 268 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 199.024 ms | 3 - 72 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8275 | 98.245 ms | 1 - 213 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 50.266 ms | 1 - 3 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® SA8775P | 51.042 ms | 1 - 214 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® SA8650P | 51.042 ms | 1 - 214 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® SA8255P | 51.042 ms | 1 - 214 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® QCS8450 | 58.248 ms | 1 - 268 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 51.634 ms | 1 - 17 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 242.962 ms | 3 - 235 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 65.959 ms | 3 - 221 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 69.809 ms | 1 - 234 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® SA7255P | 98.245 ms | 1 - 213 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® SA8295P | 58.059 ms | 1 - 219 MB | NPU
| PidNet | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 69.809 ms | 1 - 234 MB | NPU
| PidNet | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 22.237 ms | 0 - 267 MB | NPU
| PidNet | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 65.959 ms | 3 - 221 MB | NPU

## License
* The license for the original implementation of PidNet can be found
  [here](https://github.com/XuJiacong/PIDNet/blob/main/LICENSE).

## References
* [PIDNet A Real-time Semantic Segmentation Network Inspired from PID Controller Segmentation of Road Scenes](https://arxiv.org/abs/2206.02066)
* [Source Model Implementation](https://github.com/XuJiacong/PIDNet)

## 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).