tiny-dnn documentations

tiny-dnn is a header only, dependency free deep learning library written in C++. It is designed to be used in the real applications, including IoT devices and embedded systems.

• A quick introduction to tiny-dnn

• How-Tos
  • construct the network model
  • train the model
  • use/evaluate trained model
  • visualize the model
  • io
  • get/set the properties
  • tune the performance
  • handle errors
• Integrate with your application
  • Step1/3: Include tiny_dnn.h in your application
  • Step2/3: Enable C++11 options
  • Step3/3: Add include path of tiny-dnn to your build system
• Train network with your original dataset
  • 1. using opencv (image file => vec_t)
  • 2. using mnisten (image file => idx format)
  • 3. from levelDB (caffe style => [vec_t, label_t])

• Changing from v0.0.1
  • How to specify the loss and the optimizer
  • Training API for regression
  • The default mode of re-init weights

• Adding a new layer
  • Declare class
  • Implement virtual method
  • Verify backward caluculation
• How to contribute
  • Getting Started
  • Making Changes
  • Coding guides

External Links

I’m willing to update this list if your software use tiny-dnn. Please contact me at Email(see my github profile).

Official Examples

• MNIST image classification
• Cifar-10 image classification
• Deconvolutional Auto-encoder
• Importing Caffe’s model into tiny-dnn

Applications

• zhangqianhui/CnnForAndroid - A Vehicle Recognition Project using Convolutional Neural Network(CNN) in Android platform
• edgarriba/opencv_contrib (in progress) - A new opencv’s dnn module which use tiny-dnn as its backend