First of all I appreciate your response on the part 1 of issue # 3 that I encountered. Following is the query explaining the part 2 of the issue # 3 as separate issue.
Retraining and Fine-Tuning the MTL based pipeline:
While exploring the contribution of this repository, I noticed that the quickstart.ipynb guide does not provide code or guidelines for training the MTL-based pipeline with two separate datasets simultaneously. For example, I want to fine-tune the Tissue Concepts Encoder while training a Multi-Task Learning (MTL) pipeline that includes both a classification dataset and a segmentation dataset.
As this work focuses on the testing of MTL based encoder via utilizing Weakly supervised MIL heads attached to shared encoder, I wanted to explore the direction of using this MTL based shared encoder for Multi task simultaneous output generation.
Could you please provide guidance on how to use this repository to retrain and fine-tune the model with MTL, specifically when working with at least two different data sources for simultaneous output generation? For instance, I would like to use PathMNIST for classification and the SemiCol Challenge dataset for segmentation.
Any code snippets or guidelines would be greatly appreciated!
First of all I appreciate your response on the part 1 of issue # 3 that I encountered. Following is the query explaining the part 2 of the issue # 3 as separate issue.
Retraining and Fine-Tuning the MTL based pipeline:
While exploring the contribution of this repository, I noticed that the quickstart.ipynb guide does not provide code or guidelines for training the MTL-based pipeline with two separate datasets simultaneously. For example, I want to fine-tune the Tissue Concepts Encoder while training a Multi-Task Learning (MTL) pipeline that includes both a classification dataset and a segmentation dataset.
As this work focuses on the testing of MTL based encoder via utilizing Weakly supervised MIL heads attached to shared encoder, I wanted to explore the direction of using this MTL based shared encoder for Multi task simultaneous output generation.
Could you please provide guidance on how to use this repository to retrain and fine-tune the model with MTL, specifically when working with at least two different data sources for simultaneous output generation? For instance, I would like to use PathMNIST for classification and the SemiCol Challenge dataset for segmentation.
Any code snippets or guidelines would be greatly appreciated!