Joint compilation: Blake, Gao Fei On August 31, 2016, the Google team announced that it has open-sourced the latest TF-slim database for TensorFlow. It is a lightweight software package that can define, train, and evaluate models, as well as several image classification fields. The main competitive network is to test and define the model. To further advance this area, the Google team today announced the release of Inception-ResNet-v2 (a convolutional neural network - CNN), which achieved the best results in the ILSVRC image classification benchmark. Inception-ResNet-v2 is an evolution of the earlier Inception V3 model and has gained some inspiration from Microsoft's ResNet paper. Related article information can be found in our paper Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning (Inception-v4, Inception-ResNet, and the impact of residual connections on learning): Residual connections allow for shortcuts in the model, allowing researchers to successfully train deeper neural networks (which can achieve better performance), which also significantly simplifies the Inception block. Compare the two model structures, see the figure below: In the upper part of the second Inception-ResNet-v2 diagram, you can see the entire network expanded. Note that this network is considered to be deeper than the previous Inception V3. The residual blocks that are repeated in the main part of the figure have been compressed, so the entire network looks more intuitive. Also note that the inception block in the figure is simplified and contains fewer parallel towers than the previous Inception V3. As shown in the chart below, the accuracy of the Inception-ResNet-v2 architecture is higher than that of the previous best model. The chart shows that the ILSVRC 2012 image classification standard based on a single image is ranked first and fifth. Accuracy. In addition, the new model only requires twice the capacity and computing power of Inception v3. For example, although Inception v3 and Inception-ResNet-v2 are very good at identifying each dog's category, the performance of this new model is even more pronounced. For example, the old model may incorrectly identify the right picture as the Alaskan Malamute, and the new model, Inception-ResNet-v2, can accurately identify the dog's category in both pictures. Alaskan Malamute (left) and Siberian Husky (right). Image source: Wikipedia. In order for people to understand the beginning of the experiment, we will also release a new Inception-ResNet-v2 pre-training example as part of the TF-Slim image model library. Seeing the progress made in this area of ​​research on this improved model, and people are beginning to adopt this new model, and comparing its performance across multiple tasks, we are very excited about this. Do you also want to start using this new model? Let's take a look at the accompanying instructions and learn how to train, evaluate or fine-tune a network. Inception-ResNet-v2 specific code implementation process see: Https://github.com/tensorflow/models/blob/master/slim/nets/inception_resnet_v2.py PS : This article was compiled by Lei Feng Network (search "Lei Feng Network" public number) and it was compiled without permission. Via Google Research Blog DC Capacitor For Electric Furnace DC Capacitor for Electric Furnace DC Capacitor For Electric Furnace,DCMJ DC Filter Capacitors,DCMJ Pulse Capacitors,DCMJ DC Support Capacitors YANGZHOU POSITIONING TECH CO., LTD , https://www.yzpstcc.com
