My name is
Ayush Shirsat
Masters Student at Boston University
Specializing in Data Analytics
Interested in Machine Learning, Computer Vision and Robotics
My name is
Academic Projects
Analyzing EfficientNet on grading Diabetic Retinopathy
Analyzed the performance of EfficientNets on grading Diabetic Retinopathy on a scale of 0 to 4. EfficientNet models were scaled from B0 to B7 and an increase in Accuracy was observed. B7 achieved a score of 0.856 and outperformed all other single models inlcuding likes of DenseNet, ResNet and Mobile Nets.
3D Reconstruction of Telecommunication tower
Images of a telecommunication tower were taken by a drone. The tower was segmented out from the background using U-Net archictecture. Points were extracted and mapped using SIFT. 3D reconstruction was performed using Structure from Motion.
Twitter image Analysis using Google Cloud Vision
Using Tweepy API certained number of tweets were streamed. Tweets with images were stored in SQL and MongoDB database for analysis. Using Google Cloud Vision API the contents of images were displayed and stored.
Data driven enhancement of JPEG compressed images
Used Super Resolution Convolution Neural Network to enhance the quality of JPEG compressed images. The output images had better Signal to Noise ratio and reduced blocking artifact. Worked well on grayscale and RGB.
311 Haverhill Non-Emergency Request
Using QAlert API haverhill 311 non-emergency request data was collected. Used Seaborn and Matplotlib to produce data visualizations. Predicted request and response trends using regression techniques (SVM and Linear regression).