All Projects
The Crypto-UBI Project (TCUP)>
TCUP distributes a cryptocurrency basic income payment to people who have Kuwa IDs, which are “smart contracts” that live on an Ethereum blockchain programmed using Solidity. The two coolest things about TCUP are: (1) it uses the Kuwa ID system, which provides economic incentives that encourage multiple independent, rational and selfish entities to maintain a decentralized ID system that’s independent of governments, Google, Facebook or anyone else; and (2) TCUP takes a “faucet approach” in that it’s designed to distribute any cryptocurrency (e.g., bitcoin, bitcoin cash, etc.). We haven’t solved all the challenges that a crypto UBI faces, but TCUP is a substantial first step.
Visual Question Answering(VQA) using Saliency Maps>
In our project, we build upon the language and image fusion deep learning models and dataset proposed in the baseline model and introduce the idea of saliency maps to pre-process the input data to learn better feature representations of the important parts of the image. We make use of a blending mechanism to introduce the idea of a weighted dropout for feature activation instead of just using saliency activation maps. We show that our method would work slightly better than the baseline model without other attention mechanisms involved during training
Multi-task Learning(MTL) using Cross Stitch Networks>
Used Tensorflow on Python to improve accuracy on two computer vision tasks using MTL, sharing parameters between two AlexNets and introducing a joint regularisation loss.
Deep Learning - Dogs vs Cats>
Used Caffe on Python so solve a kaggle competition requiring to classify images into dogs and cats.
Person Tracker with face detection(small project) >
Used OpenCV to make a small python app to detect people using background subtraction and haars cascade classifier to detect faces.
Text Mining >
A text mining application on a question dataset to find number of questions relating to different Common Objects
Text Detection in Natural Images >
A Connected Component method derived from Stroke Width Transform to detect text regions in natural images regardless of language and font
Land Use Classifier >
A python application which used machine learning techniques to map Land use pattern of Pune, India
The MEAN Manager >
An End to End MEAN App for smooth organizational management and workflow, helpful for large teams to keep track of their work
Tarang : Spectral Library >
A spectral knowledgebase made available to sensing devices through standard framework such as Open Geo Consortium (OGC) in the form of a web service. It is a search system of field spectra for various manmade and natural materials such as urban impervious surfaces, urban vegetation, crops etc