Extract individual assets from images as transparent PNGs. Zero ML models, pure classical CV. This is used in pdf2ppt online tool pxGenius.ai
-
Updated
Mar 23, 2026 - Python
Extract individual assets from images as transparent PNGs. Zero ML models, pure classical CV. This is used in pdf2ppt online tool pxGenius.ai
PSCDL 2026 Winner — 1st Place F1 (0.40), 3rd Inference. Classical CV pipeline with circular buffer (3.4GB→186MB memory), recency filter, tests, CI/CD & Gradio demo. Python, OpenCV, NumPy.
Training-free confidence-fusion ensemble for retinal cyst segmentation in OCT B-scans, with a quantitative failure analysis showing that a collapsing retinal-band gate — not threshold choice — is the dominant error source.
Computer vision framework for multi-temporal Amazon deforestation detection using satellite imagery. Analyzes forest loss patterns, detects acceleration trends, and identifies critical periods. Features 3 detection algorithms, comprehensive testing, and validation with 20 years of real Amazon data.
Classical (non-ML) facial landmark detection in C++ using OpenCV. Detects eyes and mouth on face images using four approaches, evaluated on the MTFL dataset with Normalized Mean Error (NME).
End-to-end billiard match video analysis using purely classical computer vision: Table segmentation, ball classification, tracking, and 2D minimap projection.
Classical computer vision approach to lane detection using Canny edge detection and Hough transform
Reproducible classical computer-vision pipeline for low-light image enhancement with CLI execution and clean per-run evaluation.
Interactive image editing toolkit built with Python and OpenCV, showcasing core classical computer vision operations using real-time trackbars.
Classical machine vision system for automated geometric inspection of gear wheels. Measures 11 parameters from 2 camera views. Built as the Implement phase of a DMAIC noise reduction project for wiper motor manufacturing. Python + OpenCV. No deep learning.
A research-grade masterclass rebuilding the entire Computer Vision stack from first principles. No OpenCV, no PyTorch—just pure math and NumPy. Implements custom convolutions, Canny edge detection, eigenvalue-based Harris corners, Lucas–Kanade optical flow, and a CNN layer with full backward passes from scratch.
To associate your repository with the classical-cv topic, visit your repo's landing page and select "manage topics."