Blog

The latest news from Google AI

AutoML for large scale image classification and object detection

Thursday, November 2, 2017

Posted by Barret Zoph, Vijay Vasudevan, Jonathon Shlens and Quoc Le, Research Scientists, Google Brain TeamAutoMLImageNet Learning Transferable Architectures for Scalable Image RecognitionCOCO

We redesigned the search space so that AutoML could find the best layer which can then be stacked many times in a flexible manner to create a final network.

We performed architecture search on CIFAR-10 and transferred the best learned architecture to ImageNet image classification and COCO object detection.

“NASNet”

Our NASNet architecture is composed of two types of layers: Normal Layer (left), and Reduction Layer (right). These two layers are designed by AutoML.

Accuracies of NASNet and state-of-the-art, human-invented models at various model sizes on ImageNet image classification.

Example object detection using Faster-RCNN with NASNet.

SlimObject DetectionSpecial thanksReferencesLearning Transferable Architectures for Scalable Image RecognitionBarret Zoph, Vijay Vasudevan, Jonathon Shlens, and Quoc V. Le. Going Deeper with ConvolutionsChristian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich.Rethinking the inception architecture for computer visionChristian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Inception-v4, Inception-ResNet and the Impact of Residual Connections on LearningChristian Szegedy, Sergey Ioffe, Vincent Vanhoucke, and Alex Alemi.Squeeze-and-Excitation NetworksJie Hu, Li Shen and Gang Sun.Faster R-CNN: Towards Real-Time Object Detection with Region Proposal NetworksShaoqing Ren, Kaiming He, Ross Girshick and Jian Sun.