Detecting Facial Expressions of Comedians Captures facial landmark points using OpenCV + DLib library in Python Real-Time Face Tracking with TensorFlow.JS Captures live stream from
Mixed Reality
Augmented Reality using Markers in A-Frame + 3D Model from SketchFab Honey, I Shrunk This Guy A tiny person dancing on my desk.. Visitor from
Masterchef AI
Custom ML model to classify food / menu images, using Teachable Machines .. Demo Video Sample Output What’s Next: Try confusing the classifier by adding
Computational Music
MIDI Parser + Visualizer Parses MIDI file by extracting notes, and shows how to play the tune on piano, by highlighting corresponding keys.. Music Generation
AutoML
Model built in AutoML Vision to detect popular cartoon characters.. Model built in Auto-ML Vision to recognize popular TV / Movie Characters
Deep FAKE
Converting Still Photos into Talking Videos Here, the still pictures of famous Hollywood actors on the left automatically “lip-sync” the dialog from Shri-420 in the
Hand-Gestures
Using Hand-Gestures to Move Objects in AR / VR Instead of fixing Aruco Markers on real-world surfaces (like wall, posters, table), here I use hand-tracking
Video Segmentation
Real-Time Video Segmentation on Live Camera Stream to Extract Person from Background, and replace Background, w/o Hollywood style Green Screen Studio Setup.. Chroma-Key Effect using
Activity Recognition
Sports Activity Recognition using Transfer Learning Objective: Build a knn-classifier to recognize sports activities using Transfer Learning.. Why Transfer Learning? While pre-trained models are great
Object Detection
Object Detection using TensorFlow.JS COCO-SSD model.. Reference: TensorFlow.JS COCO-SSD model Types of Objects detected (includes most common objects like cat, dog, car, bus, truck, person,
Video Tagging
Identifying Animal / Bird Species in Wild Life Documentary Films Implemented using Mobile-Net model in TensorFlow.JS Description: analyzes video frame by frame and runs Mobile-Net
Motion Capture
Capturing body movements, gestures and actions from dance and figure skating videos using OpenPose from CMU’s panoptic studio.. Capturing Live Action (Postures, Gestures, Motion) into
Style Transfer
Applying Neural Style Transfer to produce artistic effects and filters in photos and videos.. Originals: Styled using: Implemented using PyTorch and Keras.. Try in Google
Digital Restoration
Trying Digital Restoration & Colorization of Old Black & White Films, using DeOldify, advance deep learning based implementation of Generative Adversarial Neural Networks called No-GANs..
Android TV – Streaming App
Movie streaming app installed on Android TV to watch timeless classics such as: Le Voyage Dans la Lun (1902), Metropolis (1927), Frankenstein (1910), Great Train
CINEMATRIX
Created cinematrix.in – to showcase applications of AI & Machine Learning in Multimedia & Entertainment domain.. Read the story behind this creative journey that started
3D Models
3D models built in Blender.. Available for download at – https://sketchfab.com/pamruta
3D Animation
Learning 3D Animation Basics with Blender.. Demo 1: Dancing Blocks Creates an array of 3D mesh objects, applies random color material, and wave modifier on
Fashion Advertising
Demo of Fashion Advertising created by interns from BITS Pilani, who worked in my team during Summer 2018. Displays product ads for clothing and fashion
SPORTS VIDEO ANALYTICS
Demo of Player Face Recognition created by interns from BITS Pilani, who worked in my team during Summer 2018. Uses Face Recognition to identify players
Embedded ai
Posted some cool DIY projects on Hackster at – https://www.hackster.io/pamruta
Data Viz
A cool visualization of projects posted on Kick Starter made in Kibana.. For more visualizations, visit my Tableau Public Profile at: https://public.tableau.com/app/profile/pamruta
Music Recommendation
Joined SONY‘s research team in Tokyo, Japan to work on some cutting edge technologies to create personalized music recommendations..
HUMOR ANALYSIs
Presented a paper on Humor Analysis in FRIENDS at EMNLP-2006 in Sydney. The paper analyses dialog transcripts and audio tracks in FRIENDS episodes for automatic
Amruta Purandare 