一个高度优化的轻量深度学习前向框架，使用C/C++语言开发，跨平台，支持读取Caffe模型文件，主要处理卷积神经网络。与市面上大多数移动端解决方案不同，我们的量化压缩技术不仅针对模型的权重，还涉及到输入的特征向量压缩。针对这一特性我们在模型文件和内存大小得到裁剪的同时还对框架的性能做了大量优化。(A highly optimized lightweight deep learning forward fr a mework, developed using C/C++ language, cross-platform, supported to read Caffe model files, mainly dealing with convolutional neural networks. Unlike most mobile solutions on the market, our quantitative compression techniques are not only for model weights, but also for input eigenvector compression. This feature allows us to tailor model files and memory sizes while greatly optimizing the performance of the fr a mework.)