Meysam Safarzadeh's repositories
surface_reconstruction_from_PointCloud
This project implements a neural network model for reconstructing 3D surfaces, based on the DeepSDF architecture from CVPR 2019. It uses a deep, fully-connected neural network (Decoder) to learn Signed Distance Functions (SDFs) from 3D point clouds.
SuperPixelSegmentation_using_SLIC
The project aims to segment images into rover, background, and shadow. It starts with initial segmentation using SLIC and adaptive SLIC, followed by applying a Region Adjacency Graph (RAG). To address over-segmentation, Hierarchical Merging and Normalized Cuts are used.
multimodal
This project is a multi-modal transformer based model to fuse RGB, Thermal, and depth modalities in order to predict pain intensity in 5 classes.
VIO-Using-Homography-Matrix
This application uses homography matrix, extracted from consecutive frames of a monocular camera, and a deep learning model fuses this data with input from an inertial measurement unit (IMU). This fusion is employed to accurately estimate positional changes.
Text2Shape
This project is a DL model that performs 3D shape (voxelization) generation. It takes text prompts and produces 3D shapes. The model is implemented using supervised learning and has been tested on pytorch.
drill-defect-classification
This project focuses on detecting and analyzing wear in drill bits during the drilling process. It involves studying three types of drill wear (flank, chisel, and outer corner wear) along with a healthy drill condition, using four corresponding datasets. The goal is to determine the most effective strategy for identifying drill bit wear.
rover_segmentation_using_kmeans
The project is designed to segment an image into rover, background, and shadows using a stepwise image processing algorithms. This includes color space conversion, blurring, normalization, optional contrast enhancement, feature reshaping, k-means clustering, mask merging, and morphological refinement.