MetalDAM is a metallography dataset from additive manufacturing of steels. All images were kindly provided by ArcelorMittal engineers.

Image segmentation of metallographic images is a challenging task, where features of the material that will have an impact on the final product are grouped and identified together in structures at the microscopic level imaging.

This dataset contains 42 grayscale images taken from a Scanning Electron Microscope with resolutions 1280×895 and 1024×703. These images are micrographs of steels that have been generated employing additive manufacturing techniques, and contain relevant information that can be used for quantitative and qualitative analysis of the material. An additional set of 164 unlabeled images obtained from the same materials is also provided at the same repository.

All the images in the labeled dataset have been annotated pixel-wise according to the 5 microconstituents that are present: matrix, austenite and martensite/austenite (MA), precipitates and defects. More details about class distribution are provided in the following table:

ClassRatio (%)
Martensite/Austenite (MA)8.96

The following figure shows an example of metallograph extracted from the MetalDAM dataset and its ground truth segmentation.


The article in which MetalMAD is presented is under review.


Available version of MetalDAM are downloadable at the following repository:

Publication data

June 2021


Julián Luengo Martín

Scroll Up