Yudhisteer's repositories
Point-Clouds-3D-Perception-with-Open3D
Using the KITTI dataset, we employed Open3D to visualize, downsample, segment with RANSAC, cluster via DBSCAN, create 3D bounding boxes, and perform surface reconstruction on point clouds.
UAV-Drone-Object-Tracking-using-Kalman-Filter
This project proposes the implementation of a Linear Kalman Filter from scratch to track stationary objects and individuals or animals approaching a drone's landing position, aiming to mitigate collision risks.
3D-Reconstruction-with-Uncalibrated-Stereo
The project explores 3D reconstruction using Multi-View Stereo (MVS) and Structure from Motion (SfM). We show the equations how to calibrate an uncalibrated stereo then reproject our image to 3D space.
Lane-Detection-with-Semantic-Segmentation
Lane detection using Semantic Segmentation. To be used in industries by autonomous shuttles in a controlled environment.
Predicting-the-Price-of--Anything--on-Stock-Market-Vol.1
This project is about beating the market and debugging a myth: LSTM + Stock Market = $$$.
Optical-Flow-Obstacle-Avoidance-for-UAV
In this project I aim to develop an unsupervised sense-and-avoid system for UAVs using sparse and dense optical flow.
Generating-Design-Ideas-from-Keywords
The purpose of the project is to understand a basic GAN from scratch. A WGAN was built to generate people's faces based in the Celeba Dataset. VQGAN + CLIP model was used to generate unique designs that would be used in fashion.
Pseudo-LiDARs-with-Stereo-Vision
This project focuses on harnessing the power of Pseudo-LiDARs and 3D computer vision for unmanned aerial vehicles (UAVs). By integrating Pseudo-LiDAR technology with Stereo Global Matching (SGBM) algorithms, we aim to enable UAVs to perceive their surroundings in three dimensions accurately.
Assessing-NeRF-s-Efficacy-in-3D-Model-Reconstruction-A-Comparative-Analysis-with-Blender
This project delves into NeRF basics, such as ray tracing, ray casting, and ray marching. It starts with a simple task of reconstructing a red sphere in 3D. Then, we create a custom 3D model in Blender and take images to reconstruct the model using NeRF built from scratch and train for 7 hours!
Robotic-Grasping-Detection-with-PointNet
This project focuses on training robots to grasp everyday objects accurately. We gather a unique point cloud dataset using an iPhone's LiDAR and process it with Polycam. We develop a PointNet model from scratch to perform multi-class classification and part-segmentation, guiding the robot on where to grasp objects.
Real-time-Multi-Object-Tracking-for-Rescue-Operations
This project shows the implementation of tracking algorithms like SORT and Deep SORT from scratch. The project aims to assist emergency responders in assessing the number of people at a disaster site and tracking their movements for rescue operations, especially in situations where they are being carried away by floodwaters.
Vision-Transformer-Based-Multi-Class-Classification-for-Simulated-6DoF-Robot
The project shows how to create our own large-scale synthetic dataset for multi-class classification. The goal is to create a Vision Transformer model from scratch and run it in a digital twin made in Unity with a 6 DoF robot for palletization process.
100-Data-Viz
This repositiory is about exploring fun datasets to extract insights and be able to adopt meaningful data visualization for storytelling.