Pinned Repositories
Baseline-JPEG-Implementation
JPEG Compression based on discrete cosine transform (DCT).
Classification-using-Adaboost
Classification using Adaboost with decision stumps as weak learners.
Classification-using-logistic-regression
Classification using MAP decision rule and kernelized logistic regression
Classification-using-neural-networks
Implementing a fully connected neural network with 1 and 2 hidden layers and classifying 2D synthetic data
CNN-CIFAR-10
Using CNNs to classify images in CIFAR-10 Dataset
Image-Registration
The goal of this project is to use MATLAB to implement an end-to-end image registration system, which can align image pairs to a common coordinate system. The input for performing image registration consists of two images: the original image which is considered as the reference image while the image that will be aligned with the reference image is known as the sensed image. It is the mapping of the input images with respect to the reference image. The project consists of five parts mainly concerning; point feature detection, establishing correspondences, estimating the homography, RANSAC and image warping. The resulting images are found for three different cases; Standard (unnormalized) DLT, Normalized DLT and Normalized DLT + RANSAC.
KNN_CIFAR_10
Using K Nearest Neighbors to classify images in CIFAR-10 Dataset
Linear-Regression
The goal of this assignment was to train a model to predict the price of houses using Linear regression. Housing_prices.csv contains the dataset and linear_regression.py contains code for the model
Neural-Image-Captioning-using-transfer-learning
Image captioning uses techniques from computer vision and natural language processing to predict the content of the images. To implement this, we train the model integrating both the subfields using various datasets. Datasets usually contains a plethora of images, which is both computationally expensive and time consuming to train. We propose a scheme to employ transfer learning in pre-trained models to reduce the complexity of the training phase in new datasets. This report also explains about how we used transfer learning to a model which performs a different task, to make it perform our task. Experiments on various datasets have been conducted and results have been tabulated. We analyze the accuracy of the model both qualitatively and quantitatively.
Recognition-of-Handwritten-Digits-using-the-MNIST-Dataset
The aim of this project is to implement a classification algorithm (CNN MNSIT Classifier) to recognize handwritten digits (0‐ 9). It has been shown in pattern recognition that no single classifier performs the best for all pattern classification problems consistently. An accuracy rate of 92.58% was achieved.
shruthi-menoth's Repositories
shruthi-menoth/Image-Registration
The goal of this project is to use MATLAB to implement an end-to-end image registration system, which can align image pairs to a common coordinate system. The input for performing image registration consists of two images: the original image which is considered as the reference image while the image that will be aligned with the reference image is known as the sensed image. It is the mapping of the input images with respect to the reference image. The project consists of five parts mainly concerning; point feature detection, establishing correspondences, estimating the homography, RANSAC and image warping. The resulting images are found for three different cases; Standard (unnormalized) DLT, Normalized DLT and Normalized DLT + RANSAC.
shruthi-menoth/Classification-using-neural-networks
Implementing a fully connected neural network with 1 and 2 hidden layers and classifying 2D synthetic data
shruthi-menoth/Neural-Image-Captioning-using-transfer-learning
Image captioning uses techniques from computer vision and natural language processing to predict the content of the images. To implement this, we train the model integrating both the subfields using various datasets. Datasets usually contains a plethora of images, which is both computationally expensive and time consuming to train. We propose a scheme to employ transfer learning in pre-trained models to reduce the complexity of the training phase in new datasets. This report also explains about how we used transfer learning to a model which performs a different task, to make it perform our task. Experiments on various datasets have been conducted and results have been tabulated. We analyze the accuracy of the model both qualitatively and quantitatively.
shruthi-menoth/Baseline-JPEG-Implementation
JPEG Compression based on discrete cosine transform (DCT).
shruthi-menoth/Classification-using-Adaboost
Classification using Adaboost with decision stumps as weak learners.
shruthi-menoth/Classification-using-logistic-regression
Classification using MAP decision rule and kernelized logistic regression
shruthi-menoth/CNN-CIFAR-10
Using CNNs to classify images in CIFAR-10 Dataset
shruthi-menoth/KNN_CIFAR_10
Using K Nearest Neighbors to classify images in CIFAR-10 Dataset
shruthi-menoth/Linear-Regression
The goal of this assignment was to train a model to predict the price of houses using Linear regression. Housing_prices.csv contains the dataset and linear_regression.py contains code for the model
shruthi-menoth/Recognition-of-Handwritten-Digits-using-the-MNIST-Dataset
The aim of this project is to implement a classification algorithm (CNN MNSIT Classifier) to recognize handwritten digits (0‐ 9). It has been shown in pattern recognition that no single classifier performs the best for all pattern classification problems consistently. An accuracy rate of 92.58% was achieved.
shruthi-menoth/RNN-Language-Model
Using RNN Language model to predict the next word and to generate a sentence.