Research
LightMIS offers a 131K-parameter medical image segmentation model for edge devices
The authors report Dice and IoU close to nnWNet with 98.14% fewer parameters and 94.58% fewer GFLOPs, and the full model runs about 10.5 times faster on an Arm Mali-G52 MC2 GPU. The work is a research preprint shared by its authors on Reddit.
Every outlet that ran it 1 article
r/MachineLearningCommunityLead
LightMIS: Ultra-lightweight medical image segmentation for edge devices [R]