Getting Started Installation To begin, install the keras R package from CRAN as follows: install. A Comprehensive guide to Fine-tuning Deep Learning Models in Keras (Part II) October 8, 2016 This is Part II of a 2 part series that cover fine-tuning deep learning models in Keras. The keras-vggface library provides three pre-trained VGGModels, a VGGFace1 model via model='vgg16′ (the default), and two VGGFace2 models 'resnet50' and 'senet50'. In the previous post I built a pretty good Cats vs. If you know some technical details regarding Deep Neural Networks, then you will find the Keras documentation as the best place to learn. In earlier posts, we saw the implementation of LeNet-5, AlexNet, and VGG16 which are deep convolutional neural networks. from keras_radam import RAdam RAdam (total_steps = 10000, warmup_proportion = 0. SE-ResNet-50 in Keras. js, convolution is implemented with the oft-used im2col transformation to turn it into a matrix multiply followed by reshape. 其他 2020-01-22 14:17:18 阅读次数: 0. This blog post shows the functionality and runs over a complete example using the. Deep Learning is a superpower. Similarly, we can build our own deep neural network with more than 100 layers theoretically but in reality, they are hard to train. applications. The keras-vggface library provides three pre-trained VGGModels, a VGGFace1 model via model=’vgg16′ (the default), and two VGGFace2 models ‘resnet50‘ and ‘senet50‘. # 目的 ゼロからKerasとTensorFlow(TF)を自由自在に動かせるようになる。 そのための、End to Endの作業ログ(備忘録)を残す。 ※環境はMacだが、他のOSでの汎用性を保つように意識。 ※アジャイルで執筆して. h5 vgg官方模型,适用于机器学习初学者使用。 Keras运行prisma手记(Windows) Keras运行prisma手记(Windows)曾经在ubuntu上折腾过caffe,感觉半条命都浪费在了安装中,直到遇见了keras,这是我这种新手的福音~本文不分析prisma的原理,仅仅记录我是如何通过keras运行prisma的。. But when running it with CUDA enabled: [code]python3 mnist. Last released: May 1, 2019 No project description provided. The code below loads the above model and does detection. Introduction. Training a convnet with a small dataset Having to train an image-classification model using very little data is a common situation, which you'll likely encounter in. Keras 请教一下,我在keras上用resnet50做的finetune,为什么会得到这样的结果? [图片] 这个是表示训练的时候LOSS在不断降低,但是准确度也就是比闷得稍微好一点啊。. If you are not too familiar with rospy, the comment on the first line always has to be there. I took a look at the tutorial for running keras models with tvm, and I can get that running with a single model. I only really use GEMM, but it works quite well. Some, like Keras , provide higher-level API, which makes experimentation very comfortable. The Image Classifier App is going to use Keras Deep Learning library for the image classification. Using Keras and ResNet50 pre-trained on ImageNet, we applied transfer learning to extract features from the Dogs vs. input_shape optional shape list, only to be specified if include_top is FALSE (otherwise the input shape has to be (224, 224, 3). 其他 2020-01-22 14:17:18 阅读次数: 0. 46M ResNet44 0. e ResNet10, ResNet18, ResNet50, Yolov3 and Mask RCNN. ResNet-152 in Keras. Keras: multi-label classification with ImageDataGenerator. Accuracy is measured as single-crop validation accuracy on ImageNet. Dogs classifier (with a pretty small training set) based on Keras' built-in 'ResNet50' model. In this post we'll be using the pretrained ResNet50 ImageNet weights shipped with Keras as a foundation for building a small image search engine. optional Keras tensor to use as image input for the model. The following are code examples for showing how to use keras. In this tutorial, we will discuss how to use those models. or sign in. It supports multiple back-ends, including TensorFlow, CNTK and Theano. Building ResNet in TensorFlow using Keras API Based on the plain network, we insert shortcut connections which turn the network into its counterpart residual version. Please remove it and try again. + The input layer and inference layer have to be named. My goal is to do several tests with some pre-trained standard network on the Neural Computer Stick 2, and I am working with ResNet50 by now. These models can be used for prediction, feature extraction, and fine-tuning. I only really use GEMM, but it works quite well. Keras is a high-level neural networks API developed with a focus on enabling fast experimentation. wide_resnet50_2 (pretrained=False, progress=True, **kwargs) [source] ¶ Wide ResNet-50-2 model from "Wide Residual Networks" The model is the same as ResNet except for the bottleneck number of channels which is twice larger in every block. You can vote up the examples you like or vote down the ones you don't like. SE-ResNet-50 in Keras. Keras is a popular programming framework for deep learning that simplifies the process of building deep learning applications. applications. Feature Extraction using ConvNets. input_shape optional shape list, only to be specified if include_top is FALSE (otherwise the input shape has to be (224, 224, 3). By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy. 14 Apr 2016. applications package. I wanted to find an easy way to fine tune resnet50(with imagenet weights) for grayscale images i. Transfer learning, is a research problem in machine learning that focuses on storing knowledge gained while solving one problem and applying it to a different but related problem. Visual Object Recognition in ROS Using Keras with TensorFlow I've recently gotten interested in machine learning and all of the tools that come along with that. + The model to be transferred has to be a tensorflow graph. ResNet is a powerful backbone model that is used very frequently in many computer vision tasks; ResNet uses skip connection to add the output from an earlier layer to a later. Initially, the Keras converter was developed in the project onnxmltools. Tuesday May 2, 2017. I am running on an Ubuntu 18. In this post, we take a look at what deep convolutional neural networks (convnets) really learn, and how they understand the images we feed them. preprocessing. In this tutorial, you learned how to perform online/incremental learning with Keras and the Creme machine learning library. resnet50 import ResNet50 model = ResNet50(weights=None, include_top=False,) model. First, install SystemML and other dependencies for the below demo:. Luckily, Keras Applications has a function which will return a ResNet50 as a Keras model. + The input layer and inference layer have to be named. They are stored at ~/. Keras is a wrapper for Deep Learning libraries namely Theano and TensorFlow. Keras also supplies ten well-known models, called Keras Applications, pretrained against ImageNet: Xception, VGG16, VGG19, ResNet50, InceptionV3, InceptionResNetV2, MobileNet, DenseNet, NASNet. The simplest Keras model is Sequential, which is just a linear stack of layers; other layer arrangements can be formed using the Functional model. 0dev4) from keras. utils import np_utils from matplotlib import pyplot as plt import pandas as pd #画像のサイズ指定 S…. 케라스를 사용해서 만든 모델은 모두 케라스 레이어로 이루어져 있다. Don't put anything else on the first line or else ROS won't know this is a Python script. We present a residual learning framework to ease the training of networks that are substantially deeper than those used. As an example, let's say I want to use a ResNet50 architecture to fit to my data. TensorFlow Lite for mobile and embedded devices For Production TensorFlow Extended for end-to-end ML components. After acquiring, processing, and augmenting a dataset, the next step in creating an image classifier is the construction of an appropriate model. But now, what. I'm trying to implement a simple. And now, you can create your own models on Mac using Create ML and playgrounds in Xcode 10. In Keras, models can be used as layers, and he is creating a sequential model where the first layer is the whole Resnet module. Inference is running as excpected using "label_image. Keras pre-trained models can be easily loaded as specified below − import. Deep Learning for humans. We also use the Keras vis , which is a great higher-level toolkit to visualize and debug CNNs built on Keras. 3, it should be at tf. com Abstract Deeper neural networks are more difficult to train. resnet50 import preprocess. The demo source code contains two files. Keras includes a number of deep learning models (Xception, VGG16, VGG19, ResNet50, InceptionVV3, and MobileNet) that are made available alongside pre-trained weights. The keras2onnx model converter enables users to convert Keras models into the ONNX model format. import matplotlib. I recently added this functionality into Keras' ImageDataGenerator in order to train on data that does not fit into memory. The keras-vggface library provides three pre-trained VGGModels, a VGGFace1 model via model='vgg16′ (the default), and two VGGFace2 models 'resnet50' and 'senet50'. In earlier posts, we learned about classic convolutional neural network (CNN) architectures (LeNet-5, AlexNet, VGG16, and ResNets). Two projects - Keras and tensorflow. ResNet-50 is a convolutional neural network that is trained on more than a million images from the ImageNet database. applications import imagenet_utils from PIL import Image import numpy as np import flask import io # initialize our Flask application and the Keras model app = flask. I believe it will give you some insights about the functional API and layers replacement. applications import ResNet50 from keras. It is written in Python and its biggest advantage is its ability to run on top of state-of-art deep learning libraries/frameworks such as TensorFlow, CNTK or Theano. To be added, in. To create a Keras2DML object, simply pass the keras object to the Keras2DML constructor. Available models. If you know some technical details regarding Deep Neural Networks, then you will find the Keras documentation as the best place to learn. utils import np_utils from matplotlib import pyplot as plt import pandas as pd #画像のサイズ指定 S…. If you installed keras-retinanet correctly, the train script will be installed as retinanet-train. For us to begin with, keras should be installed. ResNet50(include_top=True, weights='imagenet', input_tensor=None, input_shape=None) ResNet50 model, with weights pre-trained on ImageNet. Cats dataset. Note that the data format convention used by the model is: the one specified in your Keras config at. I don't include the top ResNet layer because I'll add my customized classification layer there. For data, we will use CIFAR10 (the standard train/test split provided by Keras) and we will resize the images to 224×224 to make them compatible with the ResNet50's. optional Keras tensor to use as image input for the model. Fine-tuning pre-trained models in Keras More to come. Accuracy is measured as single-crop validation accuracy on ImageNet. We created all the models from scratch using Keras but we didn’t train them because training such deep neural networks to require high computation cost and time. Before jumping into the problem, let's run the cell below to load the required packages. In our previous tutorial, we learned how to use models which were trained for Image Classification on the ILSVRC data. Keras models are used for prediction, feature extraction and fine tuning. keras Resnet50 imagenet. import keras model = keras. Even though ResNet is much deeper than VGG16 and VGG19, the model size is actually substantially smaller due to the usage of global average pooling rather than fully-connected layers — this reduces the model size down to 102MB for ResNet50. applications import VGG19 from keras. to hold something in front of a camera and let the computer tell you what it sees. In the remainder of this tutorial, I’ll explain what the ImageNet dataset is, and then provide Python and Keras code to classify images into 1,000 different categories using state-of-the-art network architectures. Hi Martin, There is an additional 0. set_learning_phase (0) kmodel. keras预训练模型应用(1):ResNet50进行image分类. The model might be trained using one of the many available deep learning frameworks such as Tensorflow, PyTorch, Keras, Caffe, MXNet, etc. Otherwise, it will convert it through tf. Monitoring a Keras model with TensorBoard. In this tutorial, we will present a few simple yet effective methods that you can use to build a powerful image classifier, using only very few training examples --just a few hundred or thousand pictures from each class you want to be able to recognize. resnet50_weights_tf_dim_ordering_tf_kernels_notop. 27M ResNet32 0. If you plan on training Resnet50 on real data, choose the machine type with the highest number of CPUs that you can. Here is my implementation of residual networks on Keras (on Theano). An image classification system built with transfer learning The basic technique to get transfer learning working is to get a pre-trained model (with the weights loaded) and remove final fully-connected layers from that model. Recognize images with ResNet50 model From the course: Keras is a popular programming framework for deep learning that simplifies the process of building deep learning applications. Keras pre-trained models can be easily loaded as specified below − import. Instead of providing all the functionality itself, it uses either TensorFlow or Theano behind the scenes and adds a standard, simplified programming interface on top. Tuesday May 2, 2017. It is written in Python and its biggest advantage is its ability to run on top of state-of-art deep learning libraries/frameworks such as TensorFlow, CNTK or Theano. keras are separate, with first enabling users to change between its backends and second made solely for Tensorflow…. optional Keras tensor to use as image input for the model. Multi-label classification is a useful functionality of deep neural networks. Weights are downloaded automatically when instantiating a model. Automated assessment of the number of panicles by developmental stage can provide information on the time spread of flowering and thus inform farm management. resnet50 namespace. The dataset is Stanford Dogs. Keras has a built-in function for ResNet50 pre-trained models. ResNet-152 in Keras. Multi-label classification is a useful functionality of deep neural networks. Visualizing saliency maps with ResNet50 To keep things interesting, we will conclude our smile detector experiments and actually use a pre-trained, very deep CNN to demonstrate our leopard example. Keras has this architecture at our disposal, but has the problem that, by default, the size of the images must be greater than 187 pixels, so we will define a smaller architecture. layers import Input from keras. 1% passenger_car 9. The tutorial demonstrates how to accomplish this using three of the most popular deep learning frameworks: TensorFlow, Keras, and PyTorch. ImageNet classification with Python and Keras. Keras ResNet: Building, Training & Scaling Residual Nets on Keras ResNet took the deep learning world by storm in 2015, as the first neural network that could train hundreds or thousands of layers without succumbing to the "vanishing gradient" problem. Convolution kernels directly GLSL could potentially provide speed gains I'm sure, but I can't even imagine writing it for tensors of arbitrary shape. That will ensure that your local changes will be used by the train script. OK, I Understand. Both Keras model types are now supported in the keras2onnx converter. model_selection import train_test_split import numpy as np from PIL import Image import os from glob import glob from sklearn. utils import multi_gpu_model from keras. They are from open source Python projects. ipynb, PyTorch-ResNet50. Inception, VGG16, ResNet50) out there that are helpful for overcoming sampling deficiencies; they have already been trained on many images and. Keras is a simple and powerful Python library for deep learning. resnet50_weights_tf_dim_ordering_tf_kernels_notop. when i run this tutorial that compile keras model using “relay. The network is 50 layers deep and can classify images into 1000 object categories, such as keyboard, mouse, pencil, and many animals. + The input layer and inference layer have to be named. py的执行代码与上述示例代码一致,也就是说我们可以直接运行该文档。 在存放resnet50. experimental module: Public API for tf. In earlier posts, we saw the implementation of LeNet-5, AlexNet, and VGG16 which are deep convolutional neural networks. resnet50 import ResNet50), and change the input_shape to (224,224,3) and target_size to (224,244). This net we are using (resnet50) takes tiles of Height x Width (224, 224) pixels. Image Classifier / Predictor using Keras. The winners of ILSVRC have been very generous in releasing their models to the open-source community. input_shape: optional shape list, only to be specified if include_top is FALSE (otherwise the input shape has to be (224, 224, 3). Here is my implementation of residual networks on Keras (on Theano). These models can be used for prediction, feature extraction, and fine-tuning. image import image. The architecture of ResNet50 has 4 stages as shown in the diagram below. They are from open source Python projects. The model might be trained using one of the many available deep learning frameworks such as Tensorflow, PyTorch, Keras, Caffe, MXNet, etc. py[/code] Virtual memory hits 23. As the name 'exploding' implies, during training, it causes the model's parameter to grow so large so that even a very tiny amount change in the input can cause a great update in later layers' output. This tutorial demonstrates how to use a pre-trained model for transfer learning. Fortunately, keras provides a mechanism to perform these kinds of data augmentations quickly. Keras transfer learning with ResNet50 problem. " Feb 11, 2018. 其他 2020-01-22 14:17:18 阅读次数: 0. keras预训练模型应用(1):ResNet50进行image分类. In this tutorial, we will discuss how to use those models. These models can be used for prediction, feature extraction, and fine-tuning. h5(100MB)」を ダウンロードするために 1回目は10分ほどかかる.. I saw a similar behavior when running under pytorch, so it's not very likely that it is a Keras issue. applications. Applications. I believe it will give you some insights about the functional API and layers replacement. With TensorFlow 1. We … - Selection from Hands-On Neural Networks with Keras [Book]. Given that deep learning models can take hours, days, or weeks to train, it is paramount to know how to save and load them from disk. Individually, I can get resnet50 and xception running. resnet50 for you. In our previous tutorial, we learned how to use models which were trained for Image Classification on the ILSVRC data. The simplest Keras model is Sequential, which is just a linear stack of layers; other layer arrangements can be formed using the Functional model. Keras takes away the complexities of deep learning models and provides very high level, readable API. from systemml. March 23, 2017 Motivation. But when running it with CUDA enabled: [code]python3 mnist. The keras2onnx model converter enables users to convert Keras models into the ONNX model format. 만약 중간에 텐서플로우의 레이어를 사용하고자 한다면 텐서플로우 레이어를 케라스의 Lambda 레이어로 감싸주어야 한다. The keras-vggface library provides three pre-trained VGGModels, a VGGFace1 model via model='vgg16′ (the default), and two VGGFace2 models 'resnet50' and 'senet50'. Last released: May 1, 2019 No project description provided. h5 速度快,准确率高,参数不多 50层残差网络模型,权重训练自ImageNet 该模型在Theano和TensorFlow后端均可使用,并接受channels_first和channels_last两种输入维度顺序 模型的默认输入尺寸:224x224. A Comprehensive guide to Fine-tuning Deep Learning Models in Keras (Part II) October 8, 2016 This is Part II of a 2 part series that cover fine-tuning deep learning models in Keras. •Retail Vision as a Service: A video analytics system designed for customers activity recognition using deep learning. js - Run Keras models in the browser. 0 pip install keras-resnet Copy PIP instructions. Weights are downloaded automatically when instantiating a model. Since the latest version of Keras is already supported keras. when i run this tutorial that compile keras model using “relay. applications. They are from open source Python projects. Apply the pre-trained Resnet50 deep neural network on images from the web, as a demonstration that the above works. There are many models such as AlexNet, VGGNet, Inception, ResNet, Xception and many more which we can choose from, for our own task. Image Classification using pre-trained models in Keras; Transfer Learning using pre-trained models in Keras; Fine-tuning pre-trained models in Keras; More to come. Keras is a # import the necessary packages from keras. OK, I Understand. Keras allows you to choose which lower-level library it runs on, but provides a unified API for each such backend. Keras is a wrapper for Deep Learning libraries namely Theano and TensorFlow. keras are separate, with first enabling users to change between its backends and second made solely for Tensorflow…. Similarly, we can build our own deep neural network with more than 100 layers theoretically but in reality, they are hard to train. 8888 users are registered so far. application to a Sequential model but it gives input_shape error. Both Keras model types are now supported in the keras2onnx converter. """Given a keras [model], return an equivalent model which parallelizes the computation over [n_gpus] GPUs. With TensorFlow 1. applications. Keras 上で ResNet50 を使用して分類を試してみた。 (I tried classification using ResNet50 on Keras. I am trying to understand how to add support for the TensorFlow layer FusedBatchNormV3 at the model optimizer of OpenVino. h5 速度快,准确率高,参数不多 50层残差网络模型,权重训练自ImageNet 该模型在Theano和TensorFlow后端均可使用,并接受channels_first和channels_last两种输入维度顺序 模型的默认输入尺寸:224x224. com/anujshah1003/Transfer-Learning-in-keras---custom-data This video is the continuation of Transfer learning from the first video:. 1, Keras is now at tf. Currently, Keras supports Tensorflow, CNTK and Theano. Wide ResNet¶ torchvision. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. Keras is a simple and powerful Python library for deep learning. application to a Sequential model but it gives input_shape error. 3% toyshop 8. I want to convert pretrained ResNet50 model from keras. Kerasに組み込まれているResNet50のsummaryを表示します. In the post I’d like to show how easy it is to modify the code to use an even more powerful CNN model, ‘InceptionResNetV2’. Input 0 is incompatible with layer. 6% electric_locomotive 8. After acquiring, processing, and augmenting a dataset, the next step in creating an image classifier is the construction of an appropriate model. Note that the data format convention used by the model is: the one specified in your Keras config at. Don't put anything else on the first line or else ROS won't know this is a Python script. ResNet50(include_top=True, weights='imagenet', input_tensor=None) Arguments. Keras pre-trained models can be easily loaded as specified below − import. GitHub Gist: instantly share code, notes, and snippets. 3% toyshop 8. Instead of providing all the functionality itself, it uses either TensorFlow or Theano behind the scenes and adds a standard, simplified programming interface on top. Description. utils import np_utils from matplotlib import pyplot as plt import pandas as pd #画像のサイズ指定 S…. md as there's update more up-to-date information would be there. 0 API (which is AWESOME). The core component of Keras architecture is a model. I only really use GEMM, but it works quite well. Monitoring a Keras model with TensorBoard. If the user's Keras package was installed from Keras. The example below creates a ' resnet50 ' VGGFace2 model and summarizes the shape of the inputs and outputs. Turn it into an array with image. You can vote up the examples you like or vote down the ones you don't like. Public API for tf. My previous model achieved accuracy of 98. 其他 2020-01-22 14:17:18 阅读次数: 0. If you know some technical details regarding Deep Neural Networks, then you will find the Keras documentation as the best place to learn. The example below creates a ' resnet50 ' VGGFace2 model and summarizes the shape of the inputs and outputs. Then, after preprocessing, I created a Convolutional Neural Network from scratch in Keras to classify dog breeds and then used the train/validation/test dataset to train/validate/test. Each GPU gets a slice of the input batch, applies the model on that slice. io on Slack. vgg16_weights_tf_dim_ordering_tf_kernels. TensorFlow is a lower level mathematical library for building deep neural network architectures. All the given models are available with pre-trained weights with ImageNet image database (www. Deep Learning Frameworks Speed Comparison When we want to work on Deep Learning projects, we have quite a few frameworks to choose from nowadays. ResNet50(include_top=True, weights='imagenet') model. 该模型再Theano和TensorFlow后端均可使用,并接受th和tf两种输入维度顺序. I trained the classifier with larger images (224x224, instead of 150x150). Keras: multi-label classification with ImageDataGenerator. ResNet is a pre-trained model. We are using ResNet50 model but may use other models (VGG16, VGG19, InceptionV3, etc. You received this message because you are subscribed to the Google Groups "Keras-users" group. def ResNet50 (include_top = True, weights = ' imagenet ', input_tensor = None, input_shape = None, pooling = None, classes = 1000, ** kwargs): """ Instantiates the ResNet50 architecture. from keras. This is a summary of the official Keras Documentation. I found the documentation and GitHub repo of Keras well maintained and easy to understand. The network is 50 layers deep and can classify images into 1000 object categories, such as keyboard, mouse, pencil, and many animals. The following are code examples for showing how to use keras. Instead of. GitHub Gist: instantly share code, notes, and snippets. md as there's update more up-to-date information would be there. applications import Xception # tensorflow only from keras. 0 keras Writing layers and models with tf keras (1) (0) 2019. If you plan on training Resnet50 on real data, choose the machine type with the highest number of CPUs that you can. If that isn’t a superpower, I don’t know what is. e ResNet10, ResNet18, ResNet50, Yolov3 and Mask RCNN. applications object to download the ResNet50 model in Keras format with trained parameters. applications. What is Resnet50 ?. GitHub Gist: instantly share code, notes, and snippets. Keras applications module is used to provide pre-trained model for deep neural networks. Dogs classifier (with a pretty small training set) based on Keras' built-in 'ResNet50' model. Sequential API来构建和编译一个简单的卷积神经网络 Keras 模型,用我们的 MNIST 数据集进行训练。 注意:有关构建 Keras 模型的详细训练说明,请参阅TensorFlow Keras 指南。. optional shape list, only to be specified if include_top is FALSE (otherwise the input shape has to be (224, 224, 3). Fortunately, keras provides a mechanism to perform these kinds of data augmentations quickly. resnet50 import preprocess_input, decode. Apply the pre-trained Resnet50 deep neural network on images from the web, as a demonstration that the above works. This is a summary of the official Keras Documentation. Individually, I can get resnet50 and xception running. optimizers import SGD from keras. Automated assessment of the number of panicles by developmental stage can provide information on the time spread of flowering and thus inform farm management. Pre-Trained Models in Keras Pre-trained models. The Image Classifier App is going to use Keras Deep Learning library for the image classification. These models can be used for prediction, feature extraction, and fine-tuning. Hand-Gesture Classification using Deep Convolution and Residual Neural Network (ResNet-50) with Tensorflow / Keras in Python January 20, 2018 February 14, 2018 / Sandipan Dey In this article, first an application of convolution net to classify a set of hand-sign images is going to be discussed. This article is an introductory tutorial to deploy keras models with Relay. keras Resnet50 imagenet. To install tensorflow and keras, on Ubuntu, I followed Anaconda instructions to create tensorflow environment, then used conda to install whatever was missing in that environment. Deep Learning is a superpower. optional Keras tensor to use as image input for the model. resnet50 namespace. applications import ResNet50 from keras. First, save a fixed input size training model to a Keras h5 file with both the weights and architecture. Zeiler __init__. Given that deep learning models can take hours, days and even weeks to train, it is important to know how to save and load them from disk. In the post I’d like to show how easy it is to modify the code to use an even more powerful CNN model, ‘InceptionResNetV2’. We are using ResNet50 model but may use other models (VGG16, VGG19, InceptionV3, etc.