tsegnet code #57
Description
"Thank you very much for providing the open-source code. I would like to consult about a TSegNet model effect reproduction issue. Below are the results I obtained from training your open-source model."
TSA : 0.9489540955313094 +- 0.03763659261813883
TLA : 0.454740344111534 +- 0.49740550685955026
TIR : 0.9247278692400644 +- 0.16676357242715495
score : 0.776140769627636
"The TLA metric is shockingly low. I tracked the prediction code and found that the predicted center points are inaccurate; many of the predicted center points exceed the actual number of teeth. Additionally, the categories of the center points merge with each other, resulting in segmentation where many adjacent teeth have the same label and are stuck together. I wonder if the author has run this metric and what the results were?"