Reproducible SAM-2 fine-tuning on custom bookshelf images, with scripts for local/Colab training and per-object colored masks.

Video labeler for Training and Testing Computer Vision Models Developed in Detectron2 & Pytorch

This project provides a complete pipeline to process polygon-style annotations, convert them into YOLO-compatible formats, and train a segmentation model using YOLOv8

Weakly Supervised Instance Segmentation using Class Peak Response, in CVPR 2018 (Spotlight)

PyTorch implementation of paper: Segmenting Objects by Locations

Data Collection utils for myovision Project

Instance segmentation of the cityscapes dataset.

This repository simply showcases a few results of my work under my company. I created a model that segments and localizes scratches in cars.

HTCLite implementation using PyTorch (supports MOSAIC/MixUp and RandomAugment)

Integrated Unet model for object segmentation and FasterRCNN model for object detection and instance segmentation

Deep Learning project that performs Instance Segmentation using MaskRCNN model

Localization of pickable objects from a cluttered bin of objects using a neural network trained on pure synthetic data

Train Mask-RCNN model for instance segmentation on BDD dataset

Lane detection using CNN with separate embedding and segmentation branches that yield embeddings on each lane pixel. Adopted instance segmentation approach to alleviate the issue of changing lane counts.

U-Net implementation in PyTorch. Adapted from code created for a project during my Master's degree.

Object Detection / Tracking / Instance Segmentation / Image classification / Pose Estimation

“NHIT Road Distress Detection ” states using AI&ML for detection of highway infrastructure, mainly developed pothole distress detection web and mobile app. Developed the app using SAM (segment-anything) ML model that detect the pothole overlaying a mask to view the dimensions of distress detected, stores Lat-long locations are detected

📊 Comprehensive collection of segmentation techniques - semantic, instance, panoptic. Papers, datasets and implementations.

Cross-City Building Instance Segmentation: From More Data to Diffusion-Augmentation