Model and checkpoint issues

[JcIvan]

Thank you for your outstanding contribution to the field of pathology report generation! While using your model (R2Gen) and the provided checkpoint, I noticed that the layer names in the model and the checkpoint are completely different. The model follows an encoder-decoder architecture, but the checkpoint seem to include text-transformer, visual-transformer, and text-decoder components. I tried removing unnecessary weights, but the performance deteriorated significantly.
Below is an example of the generated output:
['accompanying cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol ureter ureter ureter cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol cholesterol ']

[cpystan]

We find that the checkpoint we uploaded is wrong and should not be used for test. We are so sorry for our mistake.
We will upload a new ckpt later.

[maianhpuco]

Hi, Can you please update if the new checkpoint is uploaded?
We appreciate your contribution and want to apply your model to your work. It would help if you could provide the checkpoint.

Thank you !