TRAINING SCRIPTS

[Yingshu-Li]

I am currently conducting experiments to train a model on the MIMIC-CXR dataset and am interested in exploring different training setups. I have been experimenting with training a smaller version of UNet from scratch, and I have tested both fixing the text encoder and training it alongside the unet model. Unfortunately, the results have been suboptimal, especially when the text encoder is also being trained.

Given these challenges, I am keen to understand more about your training approach and details. Could you please share the training scripts or any specific configurations that you have used in your experiments? This would greatly help me in refining my methods and improving the results.