This SMP-UNet model was trained on a subset of the GelGenie training dataset (144 images). It builds upon the GelGenie-Universal-Dec-2023 model, extending it from 2-class to 3-class segmentation (background, bands, and wells).
The model was initialised from the GelGenie-Universal-Dec-2023 checkpoint and fine-tuned for 3-class segmentation from epoch 579 to epoch 650. Epoch 650 was selected as the final checkpoint.
Acknowledgements
Research supported with Cloud TPUs from Google's TPU Research Cloud (TRC).
