u-net-image-segmentation

Open-source ML platform for detecting deforestation, ice melt, and flooding from Sentinel-2 / Landsat imagery.

QGIS Plugin for skidtrail detection from ALS point clouds and deep learning model inference

A deep learning pipeline for semantic segmentation of cracks or defects in materials using U-Net architecture with various backbones. This project includes hyperparameter tuning, custom model architectures, and comprehensive data processing utilities.

Plataforma inteligente

An in-depth analysis of deep learning models U-Net and DeepLabV3+ in semantic segmentation, highlighting their applications in urban plaing, environmental monitoring, and geographic information systems.

Paper ECG photo → digital signal → interpretation, running entirely in the browser (YOLOv8, 1D U-Net, Pyodide, ONNX Runtime Web)

This is a joint project of Vinit Firke (myself) and Narunat Pantapalin (https://github.com/narupanta) for the Fortgeschrittene Praktikum Data Science course happened in Winter Semester 2024-25 at TU Braunschweig.

This project focuses on the detection and analysis of UHIs in Hamburg.The core of this project is to leverage multi-source data, including satellite imagery and vector data, with deep learning techniques (specifically U-Net architectures) for high-resolution UHI mapping and analysis.

End-to-end deep learning pipeline for melanoma detection using U-Net segmentation and EfficientNet classification.

U-Net for MRI segmentation.

Joint pet segmentation + 37-breed classification on Oxford-IIIT Pets: U-Net, Attention U-Net, BiFPN decoder, pretrained ResNet34-UNet (PyTorch). Suggested topics are pytorch, image-segmentation, unet, attention-unet, transfer-learning and computer-vision.

Deep learning model for automatic lung segmentation from chest X-ray images using U-Net architecture with ResNet34 encoder.

PyTorch implementation of STD-Net: Structure-Texture Demixing Network for Retinal Image Segmentation (Vessels, Optic Disc & Cup)

A project for segmenting buildings in satellite images using the U-Net architecture. Includes data preprocessing, model training, and evaluation scripts, along with preprocessed datasets and trained models.

A fully automated, high-fidelity, and multi-input GAN-based framework for precise cloud removal from satellite images with enhanced interpretability and performance across varying cloud types and terrains.

Binary building segmentation on aerial images using the Massachusetts Buildings dataset

Satellite imagery analysis using deep learning

A comprehensive deep learning project for detecting and segmenting brain diseases, particularly tumors, in MRI scans using multiple state-of-the-art architectures including U-Net and Meta's Segment Anything Model (SAM).

One of my beloved academic works that is exploring the roots and effects of an unequilibred trade-off between development expenses and gained performance has its basis residing in the current repository, which provides a full implementation of a DL model that can carry out semantic segmentation tasks while staying compact (12.2 M parm; 39 MB).