This project explores generating high-quality images using depth maps and conditioning techniques like Canny edges, leveraging Stable Diffusion and ControlNet models. It focuses on optimizing image generation with different aspect ratios, inference steps to balance speed and quality.

This library provides an easy and accessible way to read and write .pfm files in C++ by only relying on the sandard library.

Depth map estimation tool using Depth-Anything-V2. Generate accurate depth maps from images with support for both relative and metric depth measurements.

Creating a disparity map using the frames observed by a stereo-camera, after correcting epi-lines, through the use of OpenCV

A visualization tool for analyzing preprocessing results of WAI/Nerfstudio format datasets containing RGB images and corresponding depth maps to generate interactive 3D visualizations.

This repository features a customized Detection Transformer (DETR) model integrated with the NYU Depth V2 dataset, forming an end-to-end semantic segmentation pipeline. Optimized for AR applications, it utilizes depth data to enhance segmentation accuracy and enable sophisticated depth-aware scene understanding.

Program that computes depth information from a stereo camera system

I'm starting this repository to explore image depth analysis using various models. The ultimate goal is to implement Visual SLAM on the robot I previously built. This space will serve as a sandbox for experimentation—where I'll test out different depth estimation models and work on automating the robot's navigation in new environments.

Process (read and write) depth data, RGB images directly from inter realsense camera or .bag files.

Real-time depth map from drone stereo camera images with CNN

This may turn out useful if you happen to have a photometric stereo scanner or if you want to create other mappings from a given normal map.

A POC of squeezing 12-bit 400 video into 10-bit 420. Done as part of my masters thesis in May 2020.

3D-Image-Toolbox is a Python-based tool that transforms images and videos into immersive spatial experiences. Using depth-anything-v2 model, it generates depth maps from standard 2D media and converts them into side-by-side 3D formats. From spacial photos (heic format) it uses the contained depth map.

Fork of https://sourceforge.net/p/aa-project/code/ (aa3d)

This repository contains code and experiments for monocular depth estimation from single RGB images using deep convolutional neural networks. It leverages synthetic and real-world datasets, advanced data augmentation, and state-of-the-art architectures to predict depth maps for applications in robotics, autonomous driving, and augmented reality.

Implementation of algorithms for stereo vision

SinkholeNet Dataset