This model consists of an SMP-UNet trained on the entire GelGenie dataset in December 2023. It is highly robust and should produce good results for a wide variety of gel types and resolutions.
The model was trained on both the full training and validation set (total of 470 images) for 600 epochs and the final checkpoint extracted.
Its performance is comparable to the main universal model, but can sometimes be better or worse according to the gel (this model seems to perform best with higher-resolution images).
For more details on the configuration used for training, please visit https://huggingface.co/mattaq/GelGenie-Universal-Extended-Dec-2023 and check the config.toml file. Our codebase is fully open-sourced and is available here: https://github.com/mattaq31/GelGenie.