"Weld Defect detection" is a machine vision based algorithm for detecting defects (specifically weld porosity defects) in Robotic Gas Metal Arc Welding (GMAW) process in heavy equipment manufacturing. The inferencing algorithm takes a video stream capturing the welding tip as an input and reports whether (a) no weld is happening, (b) produced weld is good, or (c) produced weld is porous. The output report is provided in a csv format, summary in pdf format.