Skip to main content

Table 4 Evaluation with metrics MS-SSIM, LPIPS, and FID. Each row represents a distinct \ac{igm} trained on a varying combination of conditions time (t), treatment (trt), and simulated biomass (bm); for testing, only the input image and t are varied. MS-SSIM is reported for generations with different \(|\Delta t|\) filters: \(\text {T}_0\): identity \(|\Delta t|=0\); ST: short-term \(1\le |\Delta t|\le 10\); LT: long-term \(|\Delta t|\ge 11.\)

From: Data-driven crop growth simulation on time-varying generated images using multi-conditional generative adversarial networks

 

Train conds.

MS-SSIM (\(\uparrow\))

LPIPS (\(\downarrow\))

FID (\(\downarrow\))

t trt bm

\(\text {T}_0\)

ST

LT

ø

ø

ø

Arabidopsis

\(\checkmark\)  \(\times\)  \(\times\)

0.94

0.81

0.68

0.80

0.25

6.54

GrowliFlower

\(\checkmark\)  \(\times\)  \(\times\)

0.98

0.30

0.20

0.29

0.51

20.17

Mixed-CKA

\(\checkmark\)  \(\times\)  \(\times\)

0.99

0.23

0.22

0.30

0.46

20.44

Mixed-CKA

\(\checkmark\)  \(\checkmark\)  \(\times\)

0.97

0.25

0.23

0.31

0.47

16.26

Mixed-CKA

\(\checkmark\)  \(\checkmark\)  \(\checkmark\)

0.99

0.23

0.22

0.29

0.46

24.86

Mixed-WG\(^{1}\)

\(\checkmark\)  \(\times\)  \(\times\)

0.92

0.13

0.11

0.20

0.50

40.67

  1. \(^{1}\) Transferability check: Model trained on Mixed-CKA and applied to Mixed-WG