kishansharma3012 Goto Github PK

binary_segmentation_gmm

The aim of the project was to do binary image segmentation of grayscale images using Gaussian mixture model (GMM) and Expectation maximization (EM) algorithm. A mixture of 5 Gaussians was trained on the foreground (input rectangular region of the given image) and a mixture of 5 Gaussians was trained on the background (whole image except foreground box) using pixel wise labeling into foreground and background. After learning the foreground and background pixel intensity distribution a pixel-wise classification was done to perform final segmentation of the whole image into foreground and background.

countception

Count-Ception: Counting by Fully Convolutional Redundant Counting

deep_head_pose_estimation

Estimation of 3 euler angles to predict the pose of head from a given RGB image

dl4cv--assignment-1

dl4cv--assignment-2

dl4cv--assignment-3

face-recognition-service

Face recognition online service, allow user training it.

face_match_api

A web service for matching faces using two images, two UUIDs (from redis database) and for finding facial features

face_recognition

The world's simplest facial recognition api for Python and the command line

fashion_mnist

Training and testing different models on Zalando Fashion MNIST dataset.

histonet

kishansharma3012.github.io

My website

machine_learning_medical_imaging

object_detection_using_random_forest

The aim of the project was to do object detection and classification simultaneously using Random forest classifier. A random forest is an ensemble of multiple randomly trained decision trees. By aggregating all the predictions from different decision trees, a forest can in general yield a more robust prediction than a single tree. HOG (Histogram of Gradients) descriptors were extracted from all input images belonging to 6 different classes. A random forest was trained on HOG descriptors of the input images. Then, using sliding windows approach (using multiple aspect ratio and scale) multiple images were created from a single test image for classification. After that classification was done for each sliding window image using random forest classifier. After that non maximal suppression was used to get a single window from multiple overlapping windows around an object. Results were evaluated using IoU (intersection over union) between predicted bounding boxes and ground truth bounding boxes. A final precision-recall curve generated for different values of IoU threshold.

plant-disease-classification-

Plant disease Classification

second.pytorch

PointPillars for KITTI object detection

web-development

web development repository