Computer_Vision-CNN

Computer Vision using CNN

Introduction

Algorithm used : Convolution Nueral Network -- "Inception v3".

Task: Image recognition of “Dogs vs. Cats” using Convolutional Neural Network in python on Docker over a training dataset of 25000 images and test dataset of 12500 images.

Steps to install and run the algorithm

1. Install Docker (installing Tensorflow using Docker)

2. Download Train Data (Folder_name: task). Folder 'task' contains subfolders 'cats' and 'dogs'

3. Start Docker with local files available (linking Tensorflow Image) $ sudo docker run -it -v $HOME/tf_files:/tf_files gcr.io/tensorflow/tensorflow:latest-devel

4. Retrieve the training code $ cd /tensorflow $ git pull

5. Retrain the Inception v3 model $ python tensorflow/examples/image_retraining/retrain.py
   --bottleneck_dir=/tf_files/bottlenecks
   --how_many_training_steps 500
   --model_dir=/tf_files/inception
   --output_graph=/tf_files/retrained_graph.pb
   --output_labels=/tf_files/retrained_labels.txt
   --image_dir /tf_files/task

6. Inside Docker run the labelling script $ python /tf_files/label_image.py /tf_files/test/

7. Classifciation: A new csv file is generated "/tf_files/sample_submission.csv" having all the 'test' images with labels (1: Dog, 0: cat)

Only Testing Steps

1. Start Docker with local files available (linking Tensorflow Image) $ sudo docker run -it -v $HOME/tf_files:/tf_files gcr.io/tensorflow/tensorflow:latest-devel

2. Put all the images to be tested and labelled in the folder 'test'

3. Inside Docker run the labelling script $ python /tf_files/label_image.py /tf_files/test/

4. Classification: A new csv file is generated "/tf_files/sample_submission.csv" having all the 'test' images with labels (1: Dog, 0: cat)

5. Sort the "/tf_files/sample_submission.csv" by column 'id'

Download the whole project with dataset

https://drive.google.com/open?id=0B--hIR1DX43XY0tidm1BZnE5ODA