Skip to main content

Table 1 Dataset characteristics

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

 

Arabidopsis

GrowliFlower

MixedCrop

Mixed-CKA

Mixed-WG

# images

54,384

102,264

21,371

18,800

Observation period [d]

18

71

113

109

# times\(^{1}\) (Cond.: t)

850

12

11

10

# sequences\(^{2}\)

64

8522

2226

2212

# train sequences

40

6572

1555

1580

# val sequences

8

979

311

316

# test sequences

16

971

311

316

\(\varnothing\) images/sequence

850

12

9.60

8.50

image size [px]

256

256

256

256

GSD [mm]

0.23

3.10

5.67

5.67

diff. treatments (Cond.: trt)

\(\times\)

\(\times\)

\(\checkmark\) (76)

\(\checkmark\) (76)

sim. biomass (Cond.: bm)

\(\times\)

\(\times\)

\(\checkmark\)

\(\checkmark\)

GEM: # train images

512

1541

15,017

13,154

GEM: # val images

148

326

3177

2823

GEM: # test images

148

330

3177

2823

  1. The upper block indicates the image specifications for the image generation model, where the different conditions time (t), treatment (trt), and biomass (bm) are highlighted, and the bottom block displays the number of images used to train, validate, and test the resp. growth estimation model (GEM), which is trained independently on individual images without sequence information.
  2. \(^{1}\) The number of different time points equals the max. sequence length; for Arabidopsis, it is greater than the period because up to four images were taken per hour
  3. \(^{2}\) The number of sequences equals the number of different plants in Arabidopsis and spatially separated field patches in GrowliFlower and MixedCrop
Back to article page

Plant Methods

ISSN: 1746-4811