DiffCR learns per-layer and per-timestep token compression ratios for diffusion transformers, improving FID at similar latency relative to uniform token pruning.
Model compression and hardware acceleration for neural networks: A comprehensive survey
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2024 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Layer- and Timestep-Adaptive Differentiable Token Compression Ratios for Efficient Diffusion Transformers
DiffCR learns per-layer and per-timestep token compression ratios for diffusion transformers, improving FID at similar latency relative to uniform token pruning.