Skip to main content

Table 5 Summary of quantization works

From: A survey on GAN acceleration using memory compression techniques

Decision/work

Rep.

Func.

Granularity

Target

Module

App.

QGAN [46]

Int

EM

Network

W

G and D

Post

QMGAN [49]

Int

Uni

Network

W

G

Post

ApGAN [50] +

Int

Uni

Layer

W

G and D

During

TGAN [51] +

Int

Uni

Layer

W

G and D

During

Flexpoint [52] +

Float

Cust.

Tensor

A and W

G and D

During

GANslim [37]

Int

Uni

Network

A and W

G

Post

  1. Int integer-like, Uni uniform, A activation, + customized accelerator, Cust customized, W weight