LabeledELcells.zip

DOI - https://doi.org/10.21948/2587738

Fields

The dataset has two main folders:

img/ – contains the RGB electroluminescent images.
img/all/ – all images combined from all sources, before splitting.
img/train/ – images used for training (75% of dataset).
img/val/ – images used for validation (25% of dataset).

ann/ – contains the corresponding multi-channel binary masks (.npy format). Each mask has 4 channels, one for each defect class (Dark, Busbar, Crack, Non-cell).
ann/all/ – masks for all images in img/all/.
ann/train/ – masks for training images in img/train/.
ann/val/ – masks for validation images in img/val/.

Images in img/ and masks in ann/ have matching filenames so you can easily pair them.

Sample Data Set

Example Entry:
Image file: img/train/3_EL_18.09.2023-14-51-50_unknownID_ASU Minisample_05_57.jpg
Mask file: ann/train/3_EL_18.09.2023-14-51-50_unknownID_ASU Minisample_05_57.npy
Augmentation: None
Mask shape: (4, 256, 256)

Each of the 4 mask channels corresponds to one class in this order:
1. Channel 0 – Dark regions
2. Channel 1 – Busbars
3. Channel 2 – Cracks
4. Channel 3 – Non-cell areas

Pixels in each channel are binary (1 = present, 0 = absent), and a pixel can be active in multiple channels.
The featured image is also an example image from the dataset.