CRISTL-Project-1

The Task:

The task involved identifying the objects in the dataset, via creation of various Machine Learning models, namely- Convolutional Neural Nets, Transfer Learning Models [MobileNetV2, Densenet and VGG16]. The performance of the various models were compared and appropriate conclusions were drawn.

The dataset used is the IDD or, Indian Driving Dataset. The dataset consists of images obtained from a front facing camera attached to a car. The car was driven around Hyderabad, Bangalore cities and their outskirts. The images are mostly of 1080p resolution, but there is also some images with 720p and other resolutions. The total number of images are 10,003. The team worked on 673 images, a curtailed dataset for ease of access and handling the huge dataset in an efficient way. https://idd.insaan.iiit.ac.in/dataset/details/

Repo: https://github.com/Arunav07/CRISTL-Project-1

Link to the Gantt Chart: https://docs.google.com/document/d/18UX6mkH88T9qIU-J9HYKmB4pJGqoPlSd/edit?usp=sharing&ouid=115112343899486492342&rtpof=true&sd=true

Project Motivation

The project allows us to explore the various aspects involved while building a Machine Learning Model and gives us hands-on experience on a real life dataset. The object detection task involved will also pave a way in the future for automated cars to drive on Indian roads.

Image Dataset

Repository referred throughout the course of the project: https://github.com/BKaiwalya/Deep-Learning_Diabetic-Retinopathy-Detection

Tensorflow:

1. https://www.tensorflow.org/tutorials
2. https://www.tensorflow.org/datasets/catalog/overview
3. https://www.tensorflow.org/tutorials/keras/classification
4. https://www.tensorflow.org/tutorials/images/classification

Existing Pre-processing examples:

1. https://paperswithcode.com/dataset/idd
2. https://github.com/BKaiwalya/Deep-Learning_Diabetic-Retinopathy-Detection/blob/master/DR_InputPipeline.py

XML to CSV Conversion:

1. https://www.geeksforgeeks.org/convert-xml-to-csv-in-python/
2. https://stackoverflow.com/questions/3207219/how-do-i-list-all-files-of-a-directory

One-hot Encoding References:

1. https://en.wikipedia.org/wiki/One-hot
2. https://www.quora.com/What-is-one-hot-encoding-and-when-is-it-used-in-data-science

tf.data References:

1. https://www.tensorflow.org/guide/data
2. https://youtu.be/VFEOskzhhbc
3. https://youtu.be/kVEOCfBy9uY
4. https://cs230.stanford.edu/blog/datapipeline/#building-an-image-data-pipeline

Tf Records References:

1. https://www.tensorflow.org/tutorials/load_data/tfrecord
2. https://www.youtube.com/watch?v=oxrcZ9uUblI
3. https://towardsdatascience.com/a-practical-guide-to-tfrecords-584536bc786c

Multiple Labels Classsification References:

1. https://towardsdatascience.com/image-data-labelling-and-annotation-everything-you-need-to-know-86ede6c684b1
2. https://towardsdatascience.com/multi-label-image-classification-in-tensorflow-2-0-7d4cf8a4bc72
3. https://peltarion.com/knowledge-center/documentation/cheat-sheets/multi-label-image-classification-/-cheat-sheet

Creating label_matrix References:

1. https://numpy.org/doc/stable/reference/generated/numpy.asmatrix.html
2. https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.to_numpy.html
3. https://appdividend.com/2020/04/25/how-to-convert-python-pandas-dataframe-to-list/
4. https://www.tensorflow.org/api_docs/python/tf/convert_to_tensor

Technologies Used