Project using PyTorch in which we create custom datasets and dataloaders, train a convnext_tiny model and log it using tensorboard, do inferences and use Captum for more detailed results.

Deep learning fish classifier combining ConvNeXt-Tiny (40 species, 98.96% accuracy) with BioCLIP-2 zero-shot recognition and AI-powered habitat mapping

Mammographic images classification.

This repository contains my work on Alzheimer's Disease detection using deep learning models applied to neuroimaging data. The projects explore multiple architectures and datasets to classify Alzheimer's stages based on MRI scans.

InSwapper Detector is a production-focused deepfake detection system designed to identify faces manipulated with INSwapper. It combines face detection, RGB image analysis, frequency artifact extraction, and a multi-task ConvNeXt-Tiny model for image, batch, and video-based detection.

1st place solution for the CentraleSupélec Deep Learning Kaggle challenge "3-MD-4040 2026 ZooCAM Challenge". Plankton image classification (1.2M samples, 86 classes) using CNNs trained from scratch and a weighted logits ensemble of ResNet50, EfficientNet-B3 and ConvNeXt-Tiny with label smoothing, weighted sampling and TTA.

I built a web app for medical image analysis that allows users to upload images and receive classification results. The backend uses Spring Boot with PostgreSQL for authentication and role management, while FastAPI handles the image processing and making predictions. On the frontend, I developed a responsive ReactJS interface.

Notebooks used for my Minor Project for palm-image anemia detection with ConvNeXt baseline and ensemble experiments, which was then later used in the Major Project

A multi-modal deep learning framework for skin lesion classification on the HAM10000 dataset, combining dermoscopic images and clinical metadata using a ConvNeXt-Tiny backbone to achieve robust performance under class imbalance.

About Deep learning for ASD / Autism detection

Explainable AI-based Diabetic Retinopathy Detection using ConvNeXt-Tiny and Grad-CAM++

Breast MRI tumor classification system integrating multiple CNN models with a FastAPI backend and React frontend.

Deep learning framework for brain tumor classification using ConvNeXt, Swin Transformer, and EfficientNet, evaluated with Aquila Optimizer (AQ), baseline configurations, and Random Search (RS), with Explainable AI (Grad-CAM++, LIME, SHAP).

Classifying Brain tumor images using Late fusion of two pre trained cnn model ConvNextTiny , you can test MRI image to classify

Deep learning for ASD / Autism detection

YOLO-style object detector implemented from scratch with a custom loss function, pre-trained feature extractor, and end-to-end training pipeline on PASCAL VOC for real-time object localization and classification. Tech: Python (pytorch, scikit-learn, numpy, matplotlib, tqdm, PIL)