Who Am I?

Hi, I'm Praveen Kumar, Final year undergrad pursuing engineering from MVJ College of Engineer in Computer Science. I am a Computer Vision intern and research enthusiast working to better understand how neural networks work on object detection and instance segmentation. My expertise includes model designing, machine learning modeling, data analysis, and research paper. I have published 2 research papers. I enjoy generating new ideas and devising feasible solutions to broadly relevant problems. My colleagues would describe me as a driven, resourceful individual who maintains a positive proactive attitude when faced with adversity. Currently, I am seeking for opportunities that allow me to develop Machine Learning models and perform data analysis that benefits society. Specific fields of interest include data analytics, machine learning, deep learning, web development, and front end development.

I aim to make a difference through my creative solutions.

What I do?

Here are some of my expertise

Machine Learning

Deep Learning

Web Developement

Android Development

Cups of Tea


Pursuing Computer Science Undergraduate Full-time Course at MVJ College of Engineering.

Finished High School in Computer Science from Air Force School, ASTE.

Finished Senior Secondary from Air Force School, ASTE.


Experience / Internship

Computer Vision Intern  Jul 2019 - Present

Green Robot Machinery - VISION AI

Deep Learning applied in a real-life agricultural field. We try to find solutions to help farmers as well as find smart solutions for agricultural activities. Main tasks like deploying deep learning models on low compactible computational devices and build models to support the same.

Deep learning & Computer Vision Intern  Feb 2019 - Present

Inkers.inc, NASSCOM & NIVIDA

Learned several state of the art deep learning architecture on CNN, RNN, LSTM, GAN's and RL algorithms. Hands on Keras with projects on Image Classification, Object detection and Image Captioning. Worked on Google Colab and Tensorflow framework as back-end tools.

School Of Deep vision Nanodegree Program   Apr 2019 - Present

School Of AI

Learned neural networks such as Residual Networks, AlexNet, LeNet, VGG, CNN, RNN (LSTM and GRU), Inception (v1,v2 &v4),ResNeXt, SENET, Yolo and ENAS. Hands on projects in Image Classification, Object detection, Image captioning, Image Segmentation and Instance Segmentation and Real time deployment on above networks using Google Colab and tensorflow.



The goal of the challenge was to do as well as possible on the Image Classification problem. Tiny Imagenet has 200 classes. Each class has 500 training images, 50 validation images, and 50 test images. We have the training and validation sets with images and annotations. We have both class labels and bounding boxes as annotations; however, we are asked only to predict the class label of each image without localizing the objects.

Main author and writer for the deep learning block. The blog is a complete End-to-end Computer Vision on Keras & Tensorflow. It will provide a detailed explanation and deep understanding of all major architectures such as CNN’s, RNN’s, GAN’s, LSTM & GRU, and much more. Along with sessions, Github reporsitories will be thre for hands on experience.

Deep neural networks have attained almost human level performances over several State of the art Image classifica- tion and object detection problems. Tiny ImageNet has been here for a while and neural networks have struggled to classify them. It is one of the hardest datasets that for image classification. The validation accuracy of the existing systems max out at 61- 62% with a few shooting beyond 68-69%. These approaches are often plagued by problems of overfitting and vanishing gradients. In this paper, we present a new method to get above average validation accuracy while dodging these problems. Our model can easily adapt to the system resources availaible. We use resizing image technique which trains multiple models over different image sizes. This approach enables the model to derive more and more features as it move towards the original image size. After reaching the image size, we hyper tune other parameters such as Learning Rate, Optimizers, etc to increase the overall validation accuracy. The final validation accuracy of the model using resizing and hypertuning is 63%.

Iris dataset is the best known database to be found on pattern recognition literature. We present different clustering solutions for IRIS classification. We use and compare f1-scores of each clustering algorithm and find the possible reasons for dominance of one clustering algorithm over the others with evidences to support our claim.



Data Science Professional Certification  Feb 2019


Hands on skills on data science and machine learning. It starts with data science methodologies, and tools in data science, and then SQL and Data analysis with visualization. In the process of completion, we designed a restaurant recommender system as a capstone project using Foursquare API and developed so on geographical data.

Certificate & Credentials

My Work

Recent Work

Work 01: Nuclei Segmentation

Nuclei segemntation uses Mask-RCNN for image segmentation. We try to detect & segment all the nuclei cells from an input image.


Work 02: Skin Cancer (Lesion) disease detection

A skin lesion is a part of the skin that has an abnormal growth or appearance compared to the skin around it. Training of neural networks for automated diagnosis of pigmented skin lesions is hampered by the small size and lack of diversity of available datasets of dermatoscopic images.



Recent Blog

Nov 23, 2019 | Deep Learning

8. Introduction to Deep Learning with Computer Vision — Activation Functions

What are activation functions? How do they work? Why do we require them? What are their different types? and which one should you choose...

Oct 31, 2019 | Deep Learning

7. Introduction to Deep Learning with Computer Vision — MNIST Handwritten digit recognition hands-on

In this blog, we will learn to code and create our very first neural network. The primary task of our network will demonstrate...

Oct 17, 2019 | Deep Learning

6. Introduction to Deep Learning with Computer Vision — 3x3 is a lie! 1x1 convolutions

So far, we’ve looked at a very specific type of convolution with a kernel size of 3, more commonly called 3X3 convolutions.

Get in Touch


Marathahalli, Bangalore - 560037

Follow me on :