Embedded Files
शब्दमित्र(Shabdamitra) is an e-learning product meant for language teaching and learning. It uses Indo-Wordnet which is further augmented with audio-visual features, and grammatical properties and is presented in a learner-friendly layered format. Wordnet, an online lexical database, based on psycholinguistic principles, is built around lexical and semantic relations which are cognitive universals.
Used the concept of 'Matching the Blanks' concept for relation extraction researched by the Google Research team.
Implemented and inference on the biomedical data along with Named Entity Recognition.
Used BiLSTM-CRF architecture to detect disfluencies from the conversational text data.
Trained on Switchboard dataset with an F1 score of 94.65%.
Developed an automatic robot to detect the garbage lying on the ground using machine learning algorithms and pick up that detected garbage using a robotic arm attached to it.
Used MobileNet SSD architecture to detect the garbage bottles.
Designed an algorithm to find the 3D position of the garbage in front of the robot from the one camera view.
Developed a robotic arm with inverse kinematics to pick up the garbage and place it in the bin.
This system can distinguish garbage and valuables with more than 90% accuracy.
Researched on Credibility Examination of Human Footprint to recognize and identify humans, especially for infants.
To make a footprint dataset, a paper scanner is used in which we took 1400 images, 10 footprints per person, 5 per foot, and of 140 people.
To conquer the task of comparing test footprints to the large dataset, we developed our own CNN model which is inspired by the miniVGG model and our model can classify the footprints as per the size of the feet with the 85.5%+ test accuracy.
Developed a smart attendance system based on face recognition.
In the application of face recognition, I have implemented face detection from the image, face alignment, face embedding, the model training on my data using FaceNet architecture for the classification of the faces and recognition of the faces in real-time.
I have achieved 90%+ accuracy for 7 different person classes.
Sentimental Analysis
Sentimental Analysis
To understand the sentiments behind the users in the tweets, I am working on the sentiment analysis of Twitter's tweets using NLP.
Page updated
Google Sites
Report abuse