RohitJain88 / Deep-Learning-On-Cloud

Making a comparitive analysis of the performance of Tensorflow's object detection model on various public cloud platforms.

Deep-Learning-On-Cloud

Making a comparitive analysis of the performance of Tensorflow's object detection model on various public cloud platforms.

Getting Started

git clone https://github.com/ishanmadan1996/Deep-Learning-On-Cloud.git

Prerequisites

pip install -r requirements.txt

Steps followed

Directories:-

data/ — Will have records and csv files.
images/ — This directory will contain our dataset.
training/ — In this directory we will save our trained model.
eval/ — Will save results of evaluation on trained model.

Step 1: Generating CSV files from Images The goal is to label the image and generate train.csv and test.csv files. The detailed explanation can be found:-

https://www.youtube.com/watch?v=K_mFnvzyLvc&list=PLQVvvaa0QuDcNK5GeCQnxYnSSaar2tpku&index=3

Label the image using lablelImg. The detailed explanation can be found here. We need to convert XML into csv files which is demonstrated here:-

https://www.youtube.com/watch?v=kq2Gjv_pPe8&index=4&list=PLQVvvaa0QuDcNK5GeCQnxYnSSaar2tpku

For converting the XML to csv files we use the following script:-

xml_to_csv.py

Step 2: Generating TFRecords

In order to convert our csv file input into tfrecord (format used by tensorflow's object detection API), we can use the following script:-

generate_tfrecord.py

Step 3: Train Model

Step 4: Evaluate Model

For more detailed instructions following the steps given in below documents and links:-

Detailed_Instructions/

https://becominghuman.ai/tensorflow-object-detection-api-tutorial-training-and-evaluating-custom-object-detector-ed2594afcf73
https://pythonprogramming.net/introduction-use-tensorflow-object-detection-api-tutorial/ 

Built With

• Tensorflow Object Detection - Model used
• Python - Programming language

Authors

• Ishan Madan - Initial work - ishanmadan1996

Acknowledgments

• Special thanks to sentdex (https://www.youtube.com/user/sentdex)