This is a repository to provide information about some interesting Deep Learning Practice Problems.
The first problem we would be looking is the THE NEXAR CHALLENGE -1. It was a competition designed towards an accident free world. There are 18659 labeled training images given to find out if the traffic signal is green, red or no signal. Images are taken in the day time and in the night time. This is an amazing problem to test the capabilities of the convolutional Neural Networks (CNN).
Best place to learn CNNs is CS231n: Convolutional Neural Networks for Visual Recognition course
- Caffe I would recommend you to install Caffe locally on your machine. One of the biggest advantage of using Caffe is to have a place called Caffe Zoo, where you would get many interesting pre-trained neural network models. I understand installing Caffe is bit annoying stuff. Here are some guidelines:
- First install Anaconda2. People have reported many issues with Anaconda3 so it is better to go for Anaconda2
- OpenCV2
- I would upload the complete tutorial for this separately.
-
NVIDIA DIGITS is a deep learning platform for developing quick applications. It is an interactive GUI based on caffe platform. You can develop deep learning applications without writing single line of the code. SATYA MALLICK from Learn OpenCV has amazing tutorial for beginners for quick start of using Nvidia Digits platform. He has also created a YouTube video for the same.
-
TensorFlow is an open source Deep Learning Library by Google.
-
Keras is another deep learning library.
-
You can access AWS instance from Putty on windows
To try out your experimentations, I would recommend you to use CS231n's public AMI. If you are interested to use Jupyter Notebooks on EC2 then this Tutorial would be definitely helpful!
The experience, learnings, take aways and Open Source code and models of first four winners of the competition are made available here