Weld bead images from the literature

Dataset used in paper: "A universal convolutional neural network for pixel-level detection and monitoring of weld beads".

The dataset has ~300 weld bead images containing 2677 single-line beads collected mainly from the literature, folllowed by pixel-wise hand-annotation of the weld bead and background. They are characterized with a great variety of bead morphology, texture, contrast to background, and so on.

The ultimate goal is to train a general-purpose image-segmentation convolutional neural network (CNN), which can recognize bead regions from background across different machines and manufacturing conditions. The trained CNN is potentially deployed for weld bead monitorning and control in the general weld-based processes, such as the conventional welding and various metal depostion additive manufacturing processes.