Progressive Rising of Patch Measurement: Useful resource-Environment friendly Curriculum Studying for Dense Prediction Duties
Authors: Stefan M. Fischer, Lina Felsner, Richard Osuala, Johannes Kiechle, Daniel M. Lang, Jan C. Peeken, Julia A. Schnabel
Summary: On this work, we introduce Progressive Rising of Patch Measurement, a resource-efficient implicit curriculum studying strategy for dense prediction duties. Our curriculum strategy is outlined by rising the patch dimension throughout mannequin coaching, which step by step will increase the duty’s issue. We built-in our curriculum into the nnU-Internet framework and evaluated the methodology on all 10 duties of the Medical Segmentation Decathlon. With our strategy, we’re capable of considerably cut back runtime, computational prices, and CO2 emissions of community coaching in comparison with classical fixed patch dimension coaching. In our experiments, the curriculum strategy resulted in improved convergence. We’re capable of outperform normal nnU-Internet coaching, which is skilled with fixed patch dimension, by way of Cube Rating on 7 out of 10 MSD duties whereas solely spending roughly 50% of the unique coaching runtime. To the very best of our data, our Progressive Rising of Patch Measurement is the primary profitable employment of a sample-length curriculum within the type of patch dimension within the discipline of pc imaginative and prescient. Our code is publicly accessible at https://github.com/compai-lab/2024-miccai-fischer
