Convolutions

Invertible 1x1 Convolution

Introduced by Kingma et al. in Glow: Generative Flow with Invertible 1x1 Convolutions

The Invertible 1x1 Convolution is a type of convolution used in flow-based generative models that reverses the ordering of channels. The weight matrix is initialized as a random rotation matrix. The log-determinant of an invertible 1 × 1 convolution of a $h \times w \times c$ tensor $h$ with $c \times c$ weight matrix $\mathbf{W}$ is straightforward to compute:

$$ \log | \text{det}\left(\frac{d\text{conv2D}\left(\mathbf{h};\mathbf{W}\right)}{d\mathbf{h}}\right) | = h \cdot w \cdot \log | \text{det}\left(\mathbf{W}\right) | $$

Source: Glow: Generative Flow with Invertible 1x1 Convolutions

Papers


Paper Code Results Date Stars

Tasks


Task Papers Share
Image Dehazing 2 8.70%
Image Enhancement 2 8.70%
Speech Synthesis 2 8.70%
Pseudo Label 1 4.35%
Offline RL 1 4.35%
Text-To-Speech Synthesis 1 4.35%
Transliteration 1 4.35%
Benchmarking 1 4.35%
Molecular Docking 1 4.35%

Components


Component Type
🤖 No Components Found You can add them if they exist; e.g. Mask R-CNN uses RoIAlign

Categories