Object recognition using deep ConvNet and Caltch-256 dataset
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
Presentation
data
paper
README.md
cnn_1.py
cnn_2.py
cnn_3.py
cnn_4.py
cnn_5.py
cnn_6.py
read_data.py
read_data2.py
read_data_multiple_category.py
tester.py
vcl_setup.sh

README.md

Object-Recognition-using-ConvNet

Object recognition using deep ConvNet and Caltch-256 dataset

This project is my first foray in learning and implementing deep convolutional neural network using Tensorflow.

Need python 2.7

Packages

  1. TensorFlow
  2. Numpy
  3. Scipy
  4. Matplotlib

Folder

Data -> Contains image data (both test and train)

Files

vcl_setup.sh - Run it to setup tensorflow with GPU support and all important python packages in Ubuntu machine

read_data2.py -> Reads images for given folder and range and augments them using ritation and gaussian blur.

read_data_multiple_category.py -> Wrapper for reading images from multiple folders

cnn_*.py -> Actual CNN network. Initial code is to read the images and create label matrix. Will separate this module out later. cnn_6.py has a provision of stopping the code when the training accuracy flatlines after certein number of iterations.Please use cnn_5.py or cnn_6.py.

To test, just run cnn_6.py.

Additional Resources

References