Hugging Face's logo

qualcomm
/
MobileNet-v4

Image Classification
PyTorch
backbone
real_time
android
Model card Files
xet
 Community
MobileNet-v4
File size: 13,289 Bytes
b87ecba
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a262e8f
b87ecba
 
 
 
 
 
 
 
 
 
 
 
a262e8f
 
 
 
 
 
b87ecba
 
 
 
 
 
a262e8f
b87ecba
 
 
 
 
 
a262e8f
b87ecba
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
168
169
170
171
172
173
174
175
176
---
library_name: pytorch
license: mit
tags:
- backbone
- real_time
- android
pipeline_tag: image-classification

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/web-assets/model_demo.png)

# MobileNet-v4: Optimized for Qualcomm Devices

MobileNetV4 is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.

This is based on the implementation of MobileNet-v4 found [here](https://github.com/jaiwei98/MobileNetV4-pytorch).
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/v0.66.0/src/qai_hub_models/models/mobilenet_v4) 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 2.50, ONNX Runtime 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.50, ONNX Runtime 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/mobilenet_v4/releases/v0.66.0/mobilenet_v4-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[MobileNet-v4 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/mobilenet_v4)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.66.0/src/qai_hub_models/models/mobilenet_v4) 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 [MobileNet-v4 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.66.0/src/qai_hub_models/models/mobilenet_v4) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Input resolution: 224x224
- Model checkpoint: Imagenet
- Model size (float): 37.0 MB
- Model size (w8a8): 10.2 MB
- Number of parameters: 9.68M

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| MobileNet-v4 | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.444 ms | 0 - 43 MB | NPU
| MobileNet-v4 | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.503 ms | 0 - 39 MB | NPU
| MobileNet-v4 | ONNX | float | Snapdragon® X2 Elite | 0.379 ms | 2 - 2 MB | NPU
| MobileNet-v4 | ONNX | float | Snapdragon® X Elite | 0.823 ms | 22 - 22 MB | NPU
| MobileNet-v4 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.613 ms | 0 - 74 MB | NPU
| MobileNet-v4 | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 1.436 ms | 0 - 77 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 1.27 ms | 1 - 5 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.865 ms | 0 - 37 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® QCS8450 | 1.436 ms | 0 - 77 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 1.201 ms | 1 - 4 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 0.823 ms | 22 - 22 MB | NPU
| MobileNet-v4 | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 0.503 ms | 0 - 39 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.337 ms | 0 - 53 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.367 ms | 0 - 53 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® X2 Elite | 0.302 ms | 1 - 1 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® X Elite | 0.548 ms | 10 - 10 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.426 ms | 0 - 84 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 0.778 ms | 0 - 89 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 1.733 ms | 0 - 3 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 0.685 ms | 0 - 4 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.578 ms | 0 - 76 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® QCS8450 | 0.778 ms | 0 - 89 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 0.691 ms | 0 - 3 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 0.548 ms | 10 - 10 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 2.585 ms | 0 - 54 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 0.616 ms | 0 - 54 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.367 ms | 0 - 53 MB | NPU
| MobileNet-v4 | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.616 ms | 0 - 54 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.526 ms | 1 - 40 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.625 ms | 1 - 38 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® X2 Elite | 0.582 ms | 1 - 1 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® X Elite | 1.301 ms | 1 - 1 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 0.82 ms | 0 - 68 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 2.128 ms | 0 - 77 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 1.486 ms | 1 - 4 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.18 ms | 1 - 2 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® SA8775P | 1.589 ms | 1 - 42 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® SA8650P | 1.589 ms | 1 - 42 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® SA8255P | 1.589 ms | 1 - 42 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® QCS8450 | 2.128 ms | 0 - 77 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 1.468 ms | 1 - 3 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 1.301 ms | 1 - 1 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 0.625 ms | 1 - 38 MB | NPU
| MobileNet-v4 | QNN_DLC | float | Qualcomm® SA8295P | 2.05 ms | 0 - 38 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.276 ms | 0 - 49 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.309 ms | 0 - 51 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.347 ms | 0 - 0 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.659 ms | 0 - 0 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.419 ms | 0 - 72 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 0.77 ms | 0 - 76 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 2.097 ms | 2 - 4 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 0.634 ms | 0 - 3 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.563 ms | 0 - 2 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.739 ms | 0 - 52 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® SA8650P | 0.739 ms | 0 - 52 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® SA8255P | 0.739 ms | 0 - 52 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 0.77 ms | 0 - 76 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 0.648 ms | 0 - 2 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 0.659 ms | 0 - 0 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 2.931 ms | 0 - 49 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 0.698 ms | 0 - 50 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.309 ms | 0 - 51 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® SA7255P | 1.305 ms | 0 - 48 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Qualcomm® SA8295P | 0.998 ms | 0 - 46 MB | NPU
| MobileNet-v4 | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.698 ms | 0 - 50 MB | NPU
| MobileNet-v4 | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.517 ms | 0 - 52 MB | NPU
| MobileNet-v4 | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.624 ms | 0 - 52 MB | NPU
| MobileNet-v4 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 0.802 ms | 0 - 85 MB | NPU
| MobileNet-v4 | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 2.137 ms | 0 - 95 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 1.476 ms | 0 - 25 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.14 ms | 0 - 2 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® SA8775P | 1.597 ms | 0 - 55 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® SA8650P | 1.597 ms | 0 - 55 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® SA8255P | 1.597 ms | 0 - 55 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® QCS8450 | 2.137 ms | 0 - 95 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 1.479 ms | 0 - 25 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 0.624 ms | 0 - 52 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® SA7255P | 4.314 ms | 0 - 52 MB | NPU
| MobileNet-v4 | TFLITE | float | Qualcomm® SA8295P | 2.067 ms | 0 - 51 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.219 ms | 0 - 46 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.241 ms | 0 - 44 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.297 ms | 0 - 73 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 0.578 ms | 0 - 81 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 1.649 ms | 0 - 19 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 0.485 ms | 0 - 14 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.389 ms | 0 - 5 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® SA8775P | 0.576 ms | 0 - 50 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® SA8650P | 0.576 ms | 0 - 50 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® SA8255P | 0.576 ms | 0 - 50 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® QCS8450 | 0.578 ms | 0 - 81 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 0.489 ms | 0 - 13 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 2.332 ms | 0 - 46 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 0.529 ms | 0 - 46 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.241 ms | 0 - 44 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® SA7255P | 0.981 ms | 0 - 46 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Qualcomm® SA8295P | 0.806 ms | 0 - 43 MB | NPU
| MobileNet-v4 | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.529 ms | 0 - 46 MB | NPU

## License
* The license for the original implementation of MobileNet-v4 can be found
  [here](https://github.com/jaiwei98/MobileNetV4-pytorch/blob/main/LICENSE).

## References
* [MobileNetV4 -- Universal Models for the Mobile Ecosystem](https://arxiv.org/abs/2404.10518)
* [Source Model Implementation](https://github.com/jaiwei98/MobileNetV4-pytorch)

## 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).