Viet-Anh on Software Logo

What is: TSRUc?

SourceTransformation-based Adversarial Video Prediction on Large-Scale Data
Year2000
Data SourceCC BY-SA - https://paperswithcode.com

TSRUc, or Transformation-based Spatial Recurrent Unit c, is a modification of a ConvGRU used in the TriVD-GAN architecture for video generation.

Instead of computing the reset gate rr and resetting h_t1h\_{t−1}, the TSRUc computes the parameters of a transformation θ\theta, which we use to warp h_t1h\_{t−1}. The rest of our model is unchanged (with h^_t1\hat{h}\_{t-1} playing the role of h_th'\_{t} in cc’s update equation from ConvGRU. The TSRUc module is described by the following equations:

θ_h,x=f(h_t1,x_t)\theta\_{h,x} = f\left(h\_{t−1}, x\_{t}\right)

h^_t1=w(h_t1;θ_h,x)\hat{h}\_{t-1} = w\left(h\_{t-1}; \theta\_{h, x}\right)

c=ρ(W_c_n[h^_t1;x_t]+b_c)c = \rho\left(W\_{c} \star\_{n}\left[\hat{h}\_{t-1};x\_{t}\right] + b\_{c} \right)

u=σ(W_u_n[h_t1;x_t]+b_u)u = \sigma\left(W\_{u} \star\_{n}\left[h\_{t-1};x\_{t}\right] + b\_{u} \right)

h_t=uh_t1+(1u)ch\_{t} = u \odot h\_{t-1} + \left(1-u\right) \odot c

In these equations σ\sigma and ρ\rho are the elementwise sigmoid and ReLU functions respectively and the _n\star\_{n} represents a convolution with a kernel of size n×nn \times n. Brackets are used to represent a feature concatenation.