3D Deep Learning
Cloning the wiki
$ git clone git@github.com:Blitzman/3d-deep-learning-wiki.git
Interesting Reads and Links
- Awesome Deep Vision. A curated list of deep learning resources for computer vision.
- Deep Learning: An MIT Press book in preparation.
- Deep Learning Reading List. A list of books, papers, surveys and other readings about deep learning.
- Deep down the rabbit hole: CVPR 2015 and beyond. Insightful review of the CVPR2015, including ConvNets, Torch vs Caffe, object detectors, and ArXiv publishing.
- CVPR2015 recap and where we are going. Nice summary of the whole conference, as well as a cool summary/highlights document of some selected papers.
NVIDIA Deep Learning Course
- Introduction to Deep Learning. Video/Slides/Hands-on lab/Office hours Q&A.
- Getting started with the Caffe Framework. Video/Slides/Hands-on lab/Office hours Q&A.
Theoretical Background
Convolutional Neural Networks (CNN)
Check out the nice Stanford CS course titled CS231n: Convolutional Neural Networks for Visual Recognition. There is a nice lecture on Convolutional Neural Networks (CNNs / ConvNets) which provides an architectural overview as well as detailed explanations of the different layers and popular architectures.
A nice collection of final reports by the course students is available here. The following ones are specially remarkable:
- Tiny ImageNet Challenge - Dissection of a convolutional neural network. Explains a simple architecture and dissects it with useful insight. Besides, the figures are nice and can serve as inspiration.
- Learning 3D Object Orientations From Synthetic Images. Cool approach to mix 3D training data (which is converted to 2D) with 2D input images for object orientation detection.
- Convolutional Neural Networks for Fashion Classification and Object Detection. Multiple object detection in the same image, instead of single object per image.
Deep Neural Networks (DNN)
Frameworks
Torch
Torch is a scientific computing framework which supports a wide variety of machine learning algorithms. Characterized by an easy use thanks to its LuaJIT scripting language. It also features significant efficiency due to the underlying CUDA/C implementation.
- GitHub repository
- Convolutional Neural Network example with Lua
- Object detection with Recursive Convolutional Neural Network and Lua
Caffe
Caffe is a deep learning framework developed by the Berkeley Vision and Learning Center (BVLC). It is implemented mainly in C++ and supports GPU computing thanks to CUDA.
- GitHub repository
- arXiv / ACM MM ‘14 paper. Caffe: Convolutional Architecture for Fast Feature Embedding
- Hands-On Tutorial: Deep Learning for Vision with Caffe