Caffe2

Caffe2 (Convolutional Architecture for Fast Feature Embedding) is an open source, high-performance framework for the development of machine learning models.

Caffe2 is a popular framework due to its speed. The framework can process over 60 million images per day with a single high-performance GPU, like the Nvidia Tesla K40. The framework takes only one millisecond per image for inference and four milliseconds per image for learning.

Caffe2 supports many types of deep learning models and is specialized in image segmentation and image classification. Supported types include convolutional neural networks (CNN), recurrent neural networks (RNN), long short term memory (LSTM) and fully connected neural network designs. The framework supports Intel CPU acceleration and Nvidia GPGPU along with multi-graphics card implementations. Caffe2 will support AMD OpenCL, FPGAs, AI accelerators and CNN processors.

Yangqing Jia originally developed Caffe during his Ph.D. program at the University of Berkley, California. The current version has many contributors and is maintained by the Berkley Vision and Learning Center. The program is coded in C++ with a Python interface and is available under a BSD License. As one of its main developers, Facebook announced Caffe2 in April 2017.