Training Data and Models for the paper: Data-efficient U-Net for Segmentation of Carbide Microstructures in SEM Images of Steel Alloys

This dataset contains scanning electron microscopy (SEM) images of steel alloys, including paired secondary electron (SE2) and in-lens (InLens) channels, with corresponding binary segmentation labels. The data supports full reproduction of results presented in the referenced manuscript.

<strong>Dataset Description</strong>

<strong>Content:</strong> 13 pairs of SEM images of two reactor pressure vessel (RPV) steels:

<em>JFL</em>: IAEA reference RPV base metal steel
<em>ANP-10</em>: Western type RPV steel

<strong>Acquisition:</strong>

<em>JFL</em>: Zeiss NVision 40 microscope
<em>ANP-10</em>: Zeiss Ultra 55 microscope
Both SE and InLens detectors used simultaneously.

<strong>Resolution:</strong> 2048 × 1404 pixels per image

2048 px width corresponds to 14.3 µm (JFL) or 11.5 µm (ANP-10).

Using the dataset to reproduce the results of the manuscript

Download the zip file into the <code>data/</code> subdirectory of the code repository and extract the archive:

<pre><code class="language-bash">cd data/
unzip data.zip</code></pre>

<strong>Dataset Structure</strong>

These directories contain the relevant data for the manuscript:

<code>cloud/</code><br>
<code>├-─ preprocessed/</code><br>
<code>│   ├── hold-out/</code><br>
<code>│   ├── images/</code><br>
<code>│   └── labels/</code><br>
<code>├── processed_tiles/</code><br>
<code>│   ├── images/</code><br>
<code>│   └── labels/</code><br>
<code>├── tb_logs/</code><br>
<code>│   ├── unet_model/</code>

<strong>Preprocessed</strong>

pre-processed whole images and corresponding labels

<strong>Processed Tiles</strong>

tiled images and labels

<strong>tb_logs</strong>

trained model weights