Hi, thanks for your great work.
I have a question regarding a phenomenon I observed after fine-tuning the model on my medical imaging task.
After fine-tuning, when I use a fixed caption as input, the predicted mask sometimes shows artifacts around the boundaries, as illustrated in the attached figure. However, this issue does not happen consistently — in some cases the mask boundary looks normal, while in others these edge artifacts appear.
I am not sure what might be causing this behavior. Could this be related to the fine-tuning process, the caption input, the dataset characteristics, or the mask decoding/post-processing stage?
I would really appreciate any suggestions on possible reasons or directions for debugging this issue.
Thanks in advance!
Hi, thanks for your great work.
I have a question regarding a phenomenon I observed after fine-tuning the model on my medical imaging task.
After fine-tuning, when I use a fixed caption as input, the predicted mask sometimes shows artifacts around the boundaries, as illustrated in the attached figure. However, this issue does not happen consistently — in some cases the mask boundary looks normal, while in others these edge artifacts appear.
I am not sure what might be causing this behavior. Could this be related to the fine-tuning process, the caption input, the dataset characteristics, or the mask decoding/post-processing stage?
I would really appreciate any suggestions on possible reasons or directions for debugging this issue.
Thanks in advance!