About pretrained weights and dataset for panel segmentation.

[charles-pyj]

Hi, I am Yijun Pan, currently a upcoming senior at University of Michigan major in Data Science. Recently I am conducing a research into training multimodal models and it requires me to segment medical images. Would you kindly share pre-trained model weights or dataset you collected? It would be nice of you to provide some guidance on how you collected those data, too. Thanks!

[GaetanLepage]

Hi, Thank you for showing interest in my project. Unfortunately, although most of the pipeline is functional, I was not able to get decent enough performance on my trainings. Also, since then, several very good works have been published on this topic:

• Compound Figure Separation of Biomedical Images: Mining Large Datasets for Self-supervised Learning
• Compound Figure Separation of Biomedical Images with Side Loss
• EXSCLAIM! – An automated pipeline for the construction of labeled materials imaging datasets from literature ...

Hence, I would rather encourage you to have a look at those more advanced techniques. Of course, if you have some questions or if you want to have a discussion about my method, please do not hesitate to reach out !

Best of luck with your research.