Translucent Dataset

"Unified Generative Data augmentation for
Efficient Solar Panel Soiling Localization"

A large dataset that contains various types of solar panels and different shapes of soiling can enhance the performance of soiling segmentation models.
However, creating a new solar panel image dataset with various types of panels in the solar panel field is time-consuming and challenging work.
In this paper, we propose a unified generative framework to generate the training dataset containing various types of panels and different shapes of soiling.
We demonstrated that using the proposed dataset (called Translucent dataset) drastically enhanced the segmentation performance.

Read more >

Trnaslucent Solar Panel Soiling Image Dataset

We have proposed a new dataset, called Traslucent dataset, using unified generative augmentation.
We also demonstrate that the segmentation model trained based on the Translucent dataset was robust to the solar panel image in the wild.
the Jaccard Index results on the never-trained test set were 14.59% increases for the segmentation model trained using the Translucent Dataset compared to the model trained using the SPSI dataset.

To the best of our knowledge, this is the first solar panel soiling image dataset that contains unique soiling shapes on various types of solar panels.
You can download Translucent Dataset here