Thank you for your research.
I understand that the evaluation for DAVIS2017 is conducted using PNG files, and that no overlapping predictions should exist for a single pixel. However, since the evaluation code is not provided, it is difficult to perform an accurate assessment. Could you kindly provide the evaluation code if possible?