WnCC - Seasons of Code
Seasons of Code is a programme launched by WnCC along the lines of the Google Summer of Code. It provides one with an opprtunity to learn and participate in a variety of interesting projects under the mentorship of the very best in our institute.
List of Running Projects
- Browser Based PDF manager
- Resume Script Generator
- Physicc : A Simple Physics Engine
- Image Colorization
- Language Model Based Syntax Autocompletion in a Text Editor
- Computer vision based web app
- Cribbit Cribbit (Open for PGs Only)
- Techster Texter
- Language Detection
- Book Tracker
- ResoBin - Not the bin we deserve but the bin we need!
- Agree to disagree
- Watson (World's smartest assistant in your pocket)
- Meta Learning - Learning to Learn
- Break free of the matrix, by building one!
- Procedurally Generated Infinite Open World
- Introduction to App Development
- PAC MAN
- Introduction to Web Development
- Goal ICPC
- Traffic congestion modelling and rendering
- Tools for Data Science
- Machine Learning Based Metropolitan Air Pollution Estimation
- Audio controlled drone
- NLPlay with Transformers
- DIY FaceApp
- A Deep Dive into CNNs
- Competitive Coding
- Snake AI
- Facial Recognition App
- Gaming meets AI !!!
- R(ea)L Trader
- Computational Geometry
- Deep reinforcement learning - 2048 AI
- Reinforcement Learning to Finance
- Developing Hybrid ANN-Statistical Model for Robust Stock Market Prediction
- Astronomical Data-modelling and Interpretation
- Visual Perception for Self Driving Cars
- Convolutional Neural Networks and Applications
- Quantum Computing Algorithms
- Algorithm Visualizer
- Anime Club IITB Website using Django
- Machine Learning in Browser
Extracting potentially useful information from videos, like presence of faces, humans, specific objects, motion, tracking etc. is an indispensable part of video analysis.
Recent advancements in deep learning have demonstrated good accuracies on many of these tasks, however these models do not yet generalize well to the unseen real-world deployments.
Though the accuracy numbers for these off the shelf models have been reported and can be computed on standard existing datasets, there is currently no way to estimate how would these models perform when challenged in real-world deployments. This makes it necessary to have at least a qualitative comparison of different models against each other on such challenging, unlabelled videos.
To enable this, we envision a workbench tool (Python GUI) which will make it easy to compare any two models on a given video for a given task and “see” how they perform. It will be released as open source software. This tool will help understand the “real” strengths and weaknesses of different models for different tasks and will help give important directions to undertake future research.
To begin with we can focus on the following computer vision tasks: Motion detection, Face detection, Face recognition, Human detection, Head detection, Object detection, Tracking. A typical use of this tool will involve selecting the task, selecting a video, selecting two models/algorithms to compare and as a result seeing a side by side comparison of the analyzed video played synchronously along with other measurable parameters like time taken etc.
|Week 1||Understanding the various computer vision tasks involved|
|Week 2||Motion Detection - stand alone implementations using various alternate techniques|
|Week 3||Face Detection - stand alone implementations using various alternate off the shelf models/techniques|
|Week 4||Object detection - stand alone implementations using various alternate off the shelf models/techniques|
|Week 5||Tool design, architecture and skeleton implementation|
|Week 6||Integrating motion detection in tool|
|Week 7||Integrating face detection in tool|
|Week 8||Integrating object detection in tool|
|Week 9||Face recognition|
|Week 10||Human detection|
|Week 11||Head detection|