Back

Transformers Outperform ConvNets for Root Segmentation: A Systematic Comparison Across Nine Datasets

Smith, A. G.; Lamprinidis, S.; Seethepalli, A.; York, L. M.; Han, E.; Mohl, P.; Boulata, K.; Thorup-Kristensen, K.; Petersen, J.

2026-02-19 plant biology
10.64898/2026.02.18.706562 bioRxiv
Show abstract

Root segmentation is a fundamental yet challenging task in image-based plant phenotyping. We present the first systematic comparison of Transformer and Convolutional Neural Network (ConvNet) architectures for root segmentation, evaluating 21 architectures across nine diverse datasets and comparing pre-trained models to training from scratch. Transformer-based models significantly outperform ConvNets for segmentation accuracy and root-diameter agreement. Pre-training significantly improves mean Dice from 0.623 to 0.666 (p = 3.3 x 10-10). We also find that Transformers benefit more from pre-training than ConvNets, with Dice improvements of +0.072 versus +0.022 (p = 3.7 x 10-4), supporting the hypothesis that fine-tuned Transformers transfer more effectively across large domain gaps. Among evaluated models, MobileSAM achieved the highest Dice score while maintaining computational efficiency. Dataset choice explained far more performance variance (70.9%) than model architecture (6.7%), suggesting that data curation matters more than model selection.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

1
New Phytologist
309 papers in training set
Top 0.2%
18.0%
2
Plant Phenomics
17 papers in training set
Top 0.1%
13.9%
3
Nature Plants
84 papers in training set
Top 0.1%
12.0%
4
The Plant Journal
197 papers in training set
Top 0.4%
9.8%
50% of probability mass above
5
The Plant Phenome Journal
14 papers in training set
Top 0.1%
4.7%
6
Nature Communications
4913 papers in training set
Top 34%
4.7%
7
Applications in Plant Sciences
21 papers in training set
Top 0.1%
4.2%
8
Plant Physiology
217 papers in training set
Top 1%
3.0%
9
PLOS ONE
4510 papers in training set
Top 49%
2.0%
10
Proceedings of the National Academy of Sciences
2130 papers in training set
Top 29%
2.0%
11
Plant Communications
35 papers in training set
Top 0.6%
2.0%
12
The Plant Cell
141 papers in training set
Top 1%
1.8%
13
Genome Biology
555 papers in training set
Top 5%
1.6%
14
Scientific Reports
3102 papers in training set
Top 60%
1.6%
15
Horticulture Research
43 papers in training set
Top 1%
1.3%
16
eLife
5422 papers in training set
Top 50%
1.2%
17
Frontiers in Plant Science
240 papers in training set
Top 4%
1.1%
18
Plant Methods
39 papers in training set
Top 0.6%
0.9%
19
PLANTS, PEOPLE, PLANET
21 papers in training set
Top 0.6%
0.9%
20
Plant Biotechnology Journal
56 papers in training set
Top 1%
0.9%
21
GigaScience
172 papers in training set
Top 3%
0.8%
22
Science
429 papers in training set
Top 20%
0.7%
23
Cell Systems
167 papers in training set
Top 13%
0.7%
24
Molecular Systems Biology
142 papers in training set
Top 2%
0.6%
25
Development
440 papers in training set
Top 4%
0.6%
26
Nature Methods
336 papers in training set
Top 7%
0.6%