Segmentation, ImageNetで高精度を達成できる They propose an MDEQ that extends DEQ, which can be trained at practically infinite depths, to image-based tasks. By using equilibrium states at multiple resolutions, one model can support multiple tasks. The actual depth is only one ResBlock. And high Accuracy can be achievedthed on segmentation, ImageNet.
TL;DR
Segmentation, ImageNetで高精度を達成できる They propose an MDEQ that extends DEQ, which can be trained at practically infinite depths, to image-based tasks. By using equilibrium states at multiple resolutions, one model can support multiple tasks. The actual depth is only one ResBlock. And high Accuracy can be achievedthed on segmentation, ImageNet.
Why it matters:
Paper URL
https://arxiv.org/abs/2006.08656
Submission Dates(yyyy/mm/dd)
Authors and institutions
Methods
Results
Comments