Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Densely connected normalizing flows

About

Normalizing flows are bijective mappings between inputs and latent representations with a fully factorized distribution. They are very attractive due to exact likelihood valuation and efficient sampling. However, their effective capacity is often insufficient since the bijectivity constraint limits the model width. We address this issue by incrementally padding intermediate representations with noise. We precondition the noise in accordance with previous invertible units, which we describe as cross-unit coupling. Our invertible glow-like modules increase the model expressivity by fusing a densely connected block with Nystrom self-attention. We refer to our architecture as DenseFlow since both cross-unit and intra-module couplings rely on dense connectivity. Experiments show significant improvements due to the proposed contributions and reveal state-of-the-art density estimation under moderate computing budgets.

Matej Grci\'c, Ivan Grubi\v{s}i\'c, Sini\v{s}a \v{S}egvi\'c• 2021

Related benchmarks

TaskDatasetResultRank
Image GenerationCIFAR-10 (test)
FID34.9
471
Image GenerationCelebA 64 x 64 (test)--
203
Image GenerationCIFAR10 32x32 (test)
FID34.9
154
Image SynthesisCIFAR-10
FID34.9
79
Density EstimationCIFAR-10
bpd2.98
40
Image GenerationImageNet-32--
20
Image GenerationCIFAR-10 32x32 (train)--
20
Image GenerationImageNet 32x32 (test)--
15
Density EstimationImageNet 32 x 32
NLL (bits/dim)3.63
12
Image GenerationCelebA 64x64 (train)
Precision85.83
11
Showing 10 of 13 rows

Other info

Code

Follow for update