Hugging Face's logo

qualcomm
/
EfficientNet-Lite4

Image Classification
PyTorch
backbone
bu_auto
android
Model card Files
xet
 Community
EfficientNet-Lite4
File size: 13,722 Bytes
3f96d63
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2a3a93c
3f96d63
 
 
 
 
 
 
 
 
 
 
 
2a3a93c
 
 
 
 
 
3f96d63
 
 
 
 
 
2a3a93c
3f96d63
 
 
 
 
 
2a3a93c
3f96d63
 
 
 
 
 
 
 
 
 
 
 
 
 
2a3a93c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3f96d63
2a3a93c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3f96d63
 
 
 
 
 
 
 
 
 
 
 
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
---
library_name: pytorch
license: other
tags:
- backbone
- bu_auto
- android
pipeline_tag: image-classification

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_lite4/web-assets/model_demo.png)

# EfficientNet-Lite4: Optimized for Qualcomm Devices

EfficientNet-Lite4 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 EfficientNet-Lite4 found [here](https://github.com/RangiLyu/EfficientNet-Lite).
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/v0.58.0/src/qai_hub_models/models/efficientnet_lite4) 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 1.26.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.45, ONNX Runtime 1.26.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-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/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-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/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-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/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientnet_lite4/releases/v0.58.0/efficientnet_lite4-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[EfficientNet-Lite4 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/efficientnet_lite4)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/efficientnet_lite4) 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 [EfficientNet-Lite4 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/efficientnet_lite4) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Model checkpoint: Imagenet
- Input resolution: 300x300
- Number of parameters: 13.01M
- Model size (float): 51.9 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| EfficientNet-Lite4 | ONNX | float | Snapdragon® X2 Elite | 1.442 ms | 1 - 1 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Snapdragon® X Elite | 2.735 ms | 29 - 29 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 1.912 ms | 0 - 102 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 5.53 ms | 1 - 100 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2.633 ms | 0 - 48 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Qualcomm® QCS8450 | 5.53 ms | 1 - 100 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 4.127 ms | 1 - 4 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Snapdragon® 8 Elite Mobile | 1.497 ms | 0 - 55 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.145 ms | 0 - 57 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 1.497 ms | 0 - 55 MB | NPU
| EfficientNet-Lite4 | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 2.735 ms | 29 - 29 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® X2 Elite | 0.648 ms | 1 - 1 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® X Elite | 1.269 ms | 16 - 16 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.877 ms | 0 - 109 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 2.073 ms | 0 - 117 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 6.201 ms | 0 - 3 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.227 ms | 0 - 3 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® QCS8450 | 2.073 ms | 0 - 117 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.604 ms | 0 - 75 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.581 ms | 0 - 73 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 10.908 ms | 0 - 192 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 0.719 ms | 0 - 74 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.414 ms | 0 - 3 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 1.581 ms | 0 - 73 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.719 ms | 0 - 74 MB | NPU
| EfficientNet-Lite4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 1.269 ms | 16 - 16 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® X2 Elite | 1.864 ms | 1 - 1 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® X Elite | 3.534 ms | 1 - 1 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 2.19 ms | 0 - 103 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 6.909 ms | 0 - 98 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® QCS8275 | 15.285 ms | 1 - 53 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.166 ms | 1 - 3 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® SA8775P | 4.79 ms | 1 - 55 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® SA8650P | 4.79 ms | 1 - 55 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® SA8255P | 4.79 ms | 1 - 55 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® QCS8450 | 6.909 ms | 0 - 98 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® SA8295P | 6.439 ms | 1 - 50 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 4.511 ms | 3 - 6 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 1.687 ms | 1 - 55 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® SA7255P | 15.285 ms | 1 - 53 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.244 ms | 1 - 57 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 1.687 ms | 1 - 55 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 3.534 ms | 1 - 1 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.722 ms | 0 - 0 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® X Elite | 1.448 ms | 0 - 0 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.914 ms | 0 - 94 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 2.349 ms | 0 - 105 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 7.272 ms | 2 - 4 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® QCS8275 | 3.327 ms | 0 - 64 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.249 ms | 0 - 2 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 1.613 ms | 0 - 64 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® SA8650P | 1.613 ms | 0 - 64 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® SA8255P | 1.613 ms | 0 - 64 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 2.349 ms | 0 - 105 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.529 ms | 0 - 67 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.623 ms | 0 - 65 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® SA8295P | 2.53 ms | 0 - 61 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 11.23 ms | 0 - 185 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 0.653 ms | 0 - 62 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.431 ms | 0 - 2 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® SA7255P | 3.327 ms | 0 - 64 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 1.623 ms | 0 - 65 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.653 ms | 0 - 62 MB | NPU
| EfficientNet-Lite4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 1.448 ms | 0 - 0 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 2.193 ms | 0 - 114 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 6.671 ms | 0 - 112 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® QCS8275 | 15.283 ms | 0 - 65 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.113 ms | 0 - 3 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® SA8775P | 4.801 ms | 0 - 66 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® SA8650P | 4.801 ms | 0 - 66 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® SA8255P | 4.801 ms | 0 - 66 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® QCS8450 | 6.671 ms | 0 - 112 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® SA8295P | 6.427 ms | 0 - 48 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 4.527 ms | 0 - 32 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Snapdragon® 8 Elite Mobile | 1.686 ms | 0 - 66 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® SA7255P | 15.283 ms | 0 - 65 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.258 ms | 0 - 68 MB | NPU
| EfficientNet-Lite4 | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 1.686 ms | 0 - 66 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.66 ms | 0 - 98 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 1.739 ms | 0 - 104 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 5.834 ms | 0 - 31 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® QCS8275 | 2.815 ms | 0 - 59 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.914 ms | 0 - 15 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® SA8775P | 1.268 ms | 0 - 63 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® SA8650P | 1.268 ms | 0 - 63 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® SA8255P | 1.268 ms | 0 - 63 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® QCS8450 | 1.739 ms | 0 - 104 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.419 ms | 0 - 63 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.244 ms | 0 - 174 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® SA8295P | 2.145 ms | 0 - 56 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 10.036 ms | 0 - 178 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 0.518 ms | 0 - 57 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.089 ms | 0 - 17 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® SA7255P | 2.815 ms | 0 - 59 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 1.244 ms | 0 - 174 MB | NPU
| EfficientNet-Lite4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.518 ms | 0 - 57 MB | NPU

## License
* The license for the original implementation of EfficientNet-Lite4 can be found
  [here](https://github.com/RangiLyu/EfficientNet-Lite/blob/main/LICENSE).

## References
* [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://arxiv.org/abs/1905.11946)
* [Source Model Implementation](https://github.com/RangiLyu/EfficientNet-Lite)

## 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).