Image Classification With Tensorflow

This is a project for the Machine Learning Course by Prof.Luca Iocchi and Prof.Valsamis Ntouskos @ Sapienza University di Roma

Data set used:

Maritime Detection, Classification and tracking data set and can be downloaded from http://www.dis.uniroma1.it/~labrococo/MAR/classification.htm The training set contains images from 24 different categories of boats navigating in the City of Venice (Italy). The .rar file contains a folder for each category. The jpeg files inside the folders are named according to the date, hour, and system track number. The folder "Water" contains false positives

Let's begin!!

Install needed libraries

1.Tensorflow
https://www.tensorflow.org/install/install_linux

2.Pandas
pip install pandas

3.Numpy
pip install numpy

4.clone this repo
git clone git@github.com:adiltirur1/Image_classification.git

Getting the data sets ready

Copy the training and testing data set in to
/image_classifier/supporting_files/data_set

Let's Train the classifier

from the root directory of this repository run the command


python -m code.retrain --bottleneck_dir=supporting_files/bottlenecks --how_many_training_steps=500 --model_dir=supporting_files/models/ --summaries_dir=supporting_files/training_summaries/"${ARCHITECTURE}" --output_graph=supporting_files/retrained_graph.pb --output_labels=supporting_files/retrained_labels.txt --inception_v3 --image_dir=supporting_files/sc5

yesss!! The training is done and it shows you the estimated accuracy

Now Let us check the accuracy with the testing data set

Let us run the classifier on the testing data set and check the output

from the root directory again run the following command from the root directory again run the following command


python -m code.bulk_classify --graph=supporting_files/retrained_graph.pb

As an output of the above program, there will be a csv file created image_classification/supporting_files/results/pre_result.csv

This csv file contain the predicted output and the actual output according to the ground truth seperated by ','

Now calculate the accuracy of the bulk_prediction

navigate to
image_classification/code

Run the command


python accuracy.py

That's it, Now we have the total accuracy and an output file is created.
image_classification/supporting_files/results/final_result.csv that contains the prediction, actual output and the comparison result of actual output with the prediction