This title appears in the Scientific Report :
2017
Please use the identifier:
http://hdl.handle.net/2128/17038 in citations.
ARIGAN: Synthetic Arabidopsis Plants using Generative Adversarial Network
ARIGAN: Synthetic Arabidopsis Plants using Generative Adversarial Network
In recent years, there has been an increasing interestin image-based plant phenotyping, applying state-of-the-artmachine learning approaches to tackle challenging problems, such as leaf segmentation (a multi-instance problem)and counting. Most of these algorithms need labelled datato learn a model f...
Saved in:
Personal Name(s): | Giuffrida, Mario Valerio |
---|---|
Scharr, Hanno / Tsaftaris, Sotirios A (Corresponding author) | |
Contributing Institute: |
Pflanzenwissenschaften; IBG-2 |
Imprint: |
2017
|
Physical Description: |
2064-2071 |
Conference: | IEEE International Conference on Computer Vision Workshop, Venice (Italy), 2017-10-28 - 2017-10-28 |
Document Type: |
Contribution to a conference proceedings |
Research Program: |
Innovative Synergisms |
Link: |
Get full text OpenAccess OpenAccess |
Publikationsportal JuSER |
In recent years, there has been an increasing interestin image-based plant phenotyping, applying state-of-the-artmachine learning approaches to tackle challenging problems, such as leaf segmentation (a multi-instance problem)and counting. Most of these algorithms need labelled datato learn a model for the task at hand. Despite the recent release of a few plant phenotyping datasets, large annotatedplant image datasets for the purpose of training deep learning algorithms are lacking. One common approach to alleviate the lack of training data is dataset augmentation.Herein, we propose an alternative solution to dataset augmentation for plant phenotyping, creating artificial imagesof plants using generative neural networks. We propose theArabidopsis Rosette Image Generator (through) Adversarial Network: a deep convolutional network that is able togenerate synthetic rosette-shaped plants, inspired by DC-GAN (a recent adversarial network model using convolutional layers). Specifically, we trained the network usingA1, A2, and A4 of the CVPPP 2017 LCC dataset, containing Arabidopsis Thaliana plants. We show that our model isable to generate realistic128×128colour images of plants.We train our network conditioning on leaf count, such that itis possible to generate plants with a given number of leavessuitable, among others, for training regression based models. We propose a new Ax dataset of artificial plants images,obtained by our ARIGAN. We evaluate this new dataset using a state-of-the-art leaf counting algorithm, showing thatthe testing error is reduced when Ax is used as part of thetraining data. |