Srijan Saxena's repositories
Keypoint-regression
Trained a Convolutional Neural Network (According to paper: https://arxiv.org/pdf/1902.02394.pdf#figure.4) to detect 7 key points on a cone which is ultimately used for depth estimation using a monocamera setup on a self driving car.
NLP
The goal for this task is to take a dataset that has some labels and see if we can organize it in some unsupervised way. The dataset used in the above task is: 1)MIT Movie Corpus 2)MIT Restaurant Corpus. After this, the data has been clustered in different clusters that have similar intents using K-Means and Sentence Transformers based clustering.
ScaledYolov4
Trained a deep neural network on official SAR (Search and Rescue) dataset for drones on the IPSAR (Image Processing for Search and Rescue) project. Scaled YOLOv4 has been employed in the model that detects humans in images, which will be used on drones in missing person cases in terrain that cannot be searched by humans.
3-Tier-Architechture-
3 tier architecture
Air-Quality-Index-Prediction-
This project focuses on investigating the correlation between air quality and weather and building a prediction model based on the results of the exploratory analysis of historical climatic data of a particular region (for our project we chose Bangalore, India). This project aims to predict the air quality band for PM2.5 using present and historical climatic condition data which has been scraped from a legitimate website. After employing suitable data cleaning, data visualizing and feature engineering methods based on the observations made, the feasibility of using different machine learning techniques such as regression models and gradient boosting regressors is analyzed.
Depth-Camera_Gazebo
Extracting the video feed being published by the depth camera by creating an Image Subscriber and displaying the camera feed using OpenCV.
Facial-Detection
Facial Detection using Haar Cascade Create 2 ROS nodes, the first one should extract the webcam feed by creating an Image Subscriber. Use facial detection on the subscribed feed by implementing Haar Cascades using OpenCV. The frames with detected faces should be published to the second ROS node, which crops the bounding box (detected face) from the Image subscribed.
Museum_Management_System
A Database Management System created for museum for easier maintenance. The project is built using Visual C# (Visual Studio) for Frontend and MySql for Backend.
flownet3d
FlowNet3D: Learning Scene Flow in 3D Point Clouds (CVPR 2019)
RandLA-Net
🔥RandLA-Net in Tensorflow (CVPR 2020, Oral & IEEE TPAMI 2021)