FTCFormer replaces grid-based downsampling in vision transformers with fuzzy clustering, shared-neighbor assignment, and channel-wise token merging, reporting consistent but small accuracy improvements over TCFormer.
Title resolution pending
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
-
FTCFormer: Fuzzy Token Clustering Transformer for Image Classification
FTCFormer replaces grid-based downsampling in vision transformers with fuzzy clustering, shared-neighbor assignment, and channel-wise token merging, reporting consistent but small accuracy improvements over TCFormer.