Google Open source image classification tool Tf-slim, defining TensorFlow complex model

"Google announced today the open source TensorFlow advanced software package Tf-slim, enabling users to quickly and accurately define complex models, especially image classification tasks." This is not reminiscent of a computer vision system that Facebook last week open source "Understanding images from pixel level". In any case, there are many powerful tools in computer vision. The following is the official blog post translation. Back to "0831" to download the paper in the new smart yuan background.

Earlier this year, we released a case study of the image classification model Inception V3 on TensorFlow. The code allows the user to use the synchronous gradient descent to train the model with the IMAGENET classification database. The Inception V3 model is based on a tensorflow library called Tf-slim, which users can use to define, train, and evaluate TensorFlow models. The common abstraction provided by the Tf-slim library allows the user to define the model quickly and accurately, while ensuring that the model schema is transparent and the hyper-parameter is clear.

Since its release, Tf-slim has developed significantly, both in the network layer, the cost function, and the evaluation criteria, have added many types, training and evaluation model also has a lot of convenient routine operation. These tools allow you to run large-scale operations such as reading data in parallel or deploying models on multiple machines without worrying about the details. In addition, we have produced the Tf-slim Image Model Library, which provides definitions and training scripts for many widely used image classification models, all written using a standard database. Tf-slim and its components have been widely used within Google, and many upgrades have been integrated into the Tf.contrib.slim.

Today, we share the latest version of Tf-slim with the TF community, with highlights including:

Many new layers (such as atrous convolution and deconvolution) make the neural network architecture richer;

Support for more cost functions and evaluation indicators (e.g. Map,iou)

Deploy the runtime to make it easier to perform synchronous or asynchronous training on one or more machines

Code for defining and training widely used image classification models such as Inception, Vgg, AlexNet, ResNet

Well-trained models, these models are trained using the IMAGENET classification database, but can also be used for other computer vision tasks

ImageNet, CIFAR10 and MNIST these easy-to-use standard image databases

Use Tf-slim's GITHBU code:

Readme:https://github.com/tensorflow/models/blob/master/slim/readme.md

Instruction for use: Https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/slim