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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215

//! Models for image classification.

#![allow(clippy::upper_case_acronyms)]

use crate::download::ModelUrl;

/// Models for image classification.
#[derive(Debug, Clone)]
pub enum ImageClassification {
	/// Image classification aimed for mobile targets.
	///
	/// > MobileNet models perform image classification - they take images as input and classify the major
	/// > object in the image into a set of pre-defined classes. They are trained on ImageNet dataset which
	/// > contains images from 1000 classes. MobileNet models are also very efficient in terms of speed and
	/// > size and hence are ideal for embedded and mobile applications.
	MobileNet,
	/// Image classification, trained on ImageNet with 1000 classes.
	///
	/// > ResNet models provide very high accuracies with affordable model sizes. They are ideal for cases when
	/// > high accuracy of classification is required.
	ResNet(ResNet),
	/// A small CNN with AlexNet level accuracy on ImageNet with 50x fewer parameters.
	///
	/// > SqueezeNet is a small CNN which achieves AlexNet level accuracy on ImageNet with 50x fewer parameters.
	/// > SqueezeNet requires less communication across servers during distributed training, less bandwidth to
	/// > export a new model from the cloud to an autonomous car and more feasible to deploy on FPGAs and other
	/// > hardware with limited memory.
	SqueezeNet,
	/// Image classification, trained on ImageNet with 1000 classes.
	///
	/// > VGG models provide very high accuracies but at the cost of increased model sizes. They are ideal for
	/// > cases when high accuracy of classification is essential and there are limited constraints on model sizes.
	Vgg(Vgg),
	/// Convolutional neural network for classification, which competed in the ImageNet Large Scale Visual Recognition
	/// Challenge in 2012.
	AlexNet,
	/// Convolutional neural network for classification, which competed in the ImageNet Large Scale Visual Recognition
	/// Challenge in 2014.
	GoogleNet,
	/// Variant of AlexNet, it's the name of a convolutional neural network for classification, which competed in the
	/// ImageNet Large Scale Visual Recognition Challenge in 2012.
	CaffeNet,
	/// Convolutional neural network for detection.
	///
	/// > This model was made by transplanting the R-CNN SVM classifiers into a fc-rcnn classification layer.
	RcnnIlsvrc13,
	/// Convolutional neural network for classification.
	DenseNet121,
	/// Google's Inception
	Inception(InceptionVersion),
	/// Computationally efficient CNN architecture designed specifically for mobile devices with very limited computing
	/// power.
	ShuffleNet(ShuffleNetVersion),
	/// Deep convolutional networks for classification.
	///
	/// > This model's 4th layer has 512 maps instead of 1024 maps mentioned in the paper.
	ZFNet512,
	/// Image classification model that achieves state-of-the-art accuracy.
	///
	/// > It is designed to run on mobile CPU, GPU, and EdgeTPU devices, allowing for applications on mobile and loT,
	/// where computational resources are limited.
	EfficientNetLite4
}

#[derive(Debug, Clone)]
pub enum InceptionVersion {
	V1,
	V2
}

#[derive(Debug, Clone)]
pub enum ResNet {
	V1(ResNetV1),
	V2(ResNetV2)
}

#[derive(Debug, Clone)]
pub enum ResNetV1 {
	/// ResNet v1 with 18 layers.
	ResNet18,
	/// ResNet v1 with 34 layers.
	ResNet34,
	/// ResNet v1 with 50 layers.
	ResNet50,
	/// ResNet v1 with 101 layers.
	ResNet101,
	/// ResNet v1 with 152 layers.
	ResNet152
}

#[derive(Debug, Clone)]
pub enum ResNetV2 {
	ResNet18,
	ResNet34,
	ResNet50,
	ResNet101,
	ResNet152
}

#[derive(Debug, Clone)]
pub enum Vgg {
	/// VGG with 16 convolutional layers
	Vgg16,
	/// VGG with 16 convolutional layers, with batch normalization applied after each convolutional layer.
	///
	/// The batch normalization leads to better convergence and slightly better accuracies.
	Vgg16Bn,
	/// VGG with 19 convolutional layers
	Vgg19,
	/// VGG with 19 convolutional layers, with batch normalization applied after each convolutional layer.
	///
	/// The batch normalization leads to better convergence and slightly better accuracies.
	Vgg19Bn
}

/// Computationally efficient CNN architecture designed specifically for mobile devices with very limited computing
/// power.
#[derive(Debug, Clone)]
pub enum ShuffleNetVersion {
	/// The original ShuffleNet.
	V1,
	/// ShuffleNetV2 is an improved architecture that is the state-of-the-art in terms of speed and accuracy tradeoff
	/// used for image classification.
	V2
}

impl ModelUrl for ImageClassification {
	fn fetch_url(&self) -> &'static str {
		match self {
			ImageClassification::MobileNet => "https://github.com/onnx/models/raw/main/vision/classification/mobilenet/model/mobilenetv2-7.onnx",
			ImageClassification::SqueezeNet => "https://github.com/onnx/models/raw/main/vision/classification/squeezenet/model/squeezenet1.1-7.onnx",
			ImageClassification::Inception(version) => version.fetch_url(),
			ImageClassification::ResNet(version) => version.fetch_url(),
			ImageClassification::Vgg(variant) => variant.fetch_url(),
			ImageClassification::AlexNet => "https://github.com/onnx/models/raw/main/vision/classification/alexnet/model/bvlcalexnet-9.onnx",
			ImageClassification::GoogleNet => {
				"https://github.com/onnx/models/raw/main/vision/classification/inception_and_googlenet/googlenet/model/googlenet-9.onnx"
			}
			ImageClassification::CaffeNet => "https://github.com/onnx/models/raw/main/vision/classification/caffenet/model/caffenet-9.onnx",
			ImageClassification::RcnnIlsvrc13 => "https://github.com/onnx/models/raw/main/vision/classification/rcnn_ilsvrc13/model/rcnn-ilsvrc13-9.onnx",
			ImageClassification::DenseNet121 => "https://github.com/onnx/models/raw/main/vision/classification/densenet-121/model/densenet-9.onnx",
			ImageClassification::ShuffleNet(version) => version.fetch_url(),
			ImageClassification::ZFNet512 => "https://github.com/onnx/models/raw/main/vision/classification/zfnet-512/model/zfnet512-9.onnx",
			ImageClassification::EfficientNetLite4 => {
				"https://github.com/onnx/models/raw/main/vision/classification/efficientnet-lite4/model/efficientnet-lite4.onnx"
			}
		}
	}
}

impl ModelUrl for InceptionVersion {
	fn fetch_url(&self) -> &'static str {
		match self {
			InceptionVersion::V1 => {
				"https://github.com/onnx/models/raw/main/vision/classification/inception_and_googlenet/inception_v1/model/inception-v1-9.onnx"
			}
			InceptionVersion::V2 => {
				"https://github.com/onnx/models/raw/main/vision/classification/inception_and_googlenet/inception_v2/model/inception-v2-9.onnx"
			}
		}
	}
}

impl ModelUrl for ResNet {
	fn fetch_url(&self) -> &'static str {
		match self {
			ResNet::V1(variant) => variant.fetch_url(),
			ResNet::V2(variant) => variant.fetch_url()
		}
	}
}

impl ModelUrl for ResNetV1 {
	fn fetch_url(&self) -> &'static str {
		match self {
			ResNetV1::ResNet18 => "https://github.com/onnx/models/raw/main/vision/classification/resnet/model/resnet18-v1-7.onnx",
			ResNetV1::ResNet34 => "https://github.com/onnx/models/raw/main/vision/classification/resnet/model/resnet34-v1-7.onnx",
			ResNetV1::ResNet50 => "https://github.com/onnx/models/raw/main/vision/classification/resnet/model/resnet50-v1-7.onnx",
			ResNetV1::ResNet101 => "https://github.com/onnx/models/raw/main/vision/classification/resnet/model/resnet101-v1-7.onnx",
			ResNetV1::ResNet152 => "https://github.com/onnx/models/raw/main/vision/classification/resnet/model/resnet152-v1-7.onnx"
		}
	}
}

impl ModelUrl for ResNetV2 {
	fn fetch_url(&self) -> &'static str {
		match self {
			ResNetV2::ResNet18 => "https://github.com/onnx/models/raw/main/vision/classification/resnet/model/resnet18-v2-7.onnx",
			ResNetV2::ResNet34 => "https://github.com/onnx/models/raw/main/vision/classification/resnet/model/resnet34-v2-7.onnx",
			ResNetV2::ResNet50 => "https://github.com/onnx/models/raw/main/vision/classification/resnet/model/resnet50-v2-7.onnx",
			ResNetV2::ResNet101 => "https://github.com/onnx/models/raw/main/vision/classification/resnet/model/resnet101-v2-7.onnx",
			ResNetV2::ResNet152 => "https://github.com/onnx/models/raw/main/vision/classification/resnet/model/resnet152-v2-7.onnx"
		}
	}
}

impl ModelUrl for Vgg {
	fn fetch_url(&self) -> &'static str {
		match self {
			Vgg::Vgg16 => "https://github.com/onnx/models/raw/main/vision/classification/vgg/model/vgg16-7.onnx",
			Vgg::Vgg16Bn => "https://github.com/onnx/models/raw/main/vision/classification/vgg/model/vgg16-bn-7.onnx",
			Vgg::Vgg19 => "https://github.com/onnx/models/raw/main/vision/classification/vgg/model/vgg19-7.onnx",
			Vgg::Vgg19Bn => "https://github.com/onnx/models/raw/main/vision/classification/vgg/model/vgg19-bn-7.onnx"
		}
	}
}

impl ModelUrl for ShuffleNetVersion {
	fn fetch_url(&self) -> &'static str {
		match self {
			ShuffleNetVersion::V1 => "https://github.com/onnx/models/raw/main/vision/classification/shufflenet/model/shufflenet-9.onnx",
			ShuffleNetVersion::V2 => "https://github.com/onnx/models/raw/main/vision/classification/shufflenet/model/shufflenet-v2-10.onnx"
		}
	}
}