The Effects of Synthetic Data and Label Distribution on Canola Branch Counting

2026-07-10

AI Summary

This study addresses the high annotation cost and low efficiency in rapeseed branch counting by leveraging calibrated L-systems to generate synthetic images with precise labels. It systematically investigates the impact of the ratio between synthetic and real data, as well as label distribution, on the performance of a ResNet-18 model for plant phenotyping. The work quantifies, for the first time, the independent contributions of these two factors and introduces a label optimization strategy based on interpolation according to the real-data distribution followed by Gaussian smoothing. Experimental results demonstrate that an optimal synthetic-to-real data ratio of 1:7 reduces the mean absolute error by 7.6%, and further integration of Gaussian smoothing lowers the error to 0.912, representing a 14.7% improvement over using real data alone.

Abstract

Collecting annotated plant images for automated phenotyping is often slow and expensive. Plant models simulating growth and development can generate unlimited synthetic images with exact labels. However, previous work has established that whether incorporating synthetic data improves performance depends on the ratio of synthetic to real images and the label distribution of the synthetic dataset. To systematically quantify both factors, we train ResNet-18 models on a canola branch-counting task using a calibrated L-system plant model. We vary each factor independently. Synthetic-to-real ratios of 1:5 to 1:22 broadly improve performance; the best ratio (1:7) reduces mean absolute difference by 7.6% over real-only training. For label distribution, a uniform synthetic distribution is strongly suboptimal (abs. diff. of approximately 1.70); interpolating 90% toward the real distribution yields abs. diff. 0.927, whereas Gaussian smoothing of the real label distribution yields the best overall result (abs. diff. 0.912, a 14.7% improvement over real-only). A minimum of 10 synthetic images per label offers a simpler alternative with modest gains, while 100 per label over-corrects and hurts performance.

Problem

• synthetic data
• label distribution
• canola branch counting
• automated phenotyping
• data ratio

Innovation

• synthetic data
• label distribution
• L-system modeling
• phenotyping
• data augmentation