A systematic study shows that tuning only the late layers of SAM's vision transformer gives the best memory-accuracy trade-off for biomedical segmentation.
Segment anything for microscopy.Nature Methods, 22(3):579–591, February 2025a
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Parameter Efficient Fine-Tuning of Segment Anything Model for Biomedical Imaging
A systematic study shows that tuning only the late layers of SAM's vision transformer gives the best memory-accuracy trade-off for biomedical segmentation.