iFAN adds a mask-quality ranking loss and a cross-layer self-distillation loss to mask transformer training, improving panoptic, instance, and semantic segmentation by about one point with no inference overhead.
Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
iFAN: Inference-Aware Learning for Plain Mask Transformers
iFAN adds a mask-quality ranking loss and a cross-layer self-distillation loss to mask transformer training, improving panoptic, instance, and semantic segmentation by about one point with no inference overhead.