[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82432-en":3,"doc-seo-82432-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},82432,7971461741311,"Ophelia","https://ap-avatar.wpscdn.com/avatar/74000253aff267980c6?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779345379180704826",8,"Research & Report","The Effects of Synthetic Data and Label Distribution on Canola Branch Counting","Collecting annotated plant images for automated phenotyping is slow and expensive, while calibrated growth models can generate unlimited synthetic images with exact branch-count labels. To quantify how synthetic-to-real ratio and synthetic label distribution affect performance, ResNet-18 is trained on a canola branch-counting task using a calibrated L-system model. Ratios from 1:5 to 1:22 help, with 1:7 lowering mean absolute difference by 7.6%; for distributions, Gaussian smoothing of real labels gives the best overall result.","The Effects of Synthetic Data and Label Distribution on Canola Branch Counting  \nAmirsalar Darvishpour, Mikolaj Cieslak, and Adam Runions  \nDepartment of Computer Science, University of Calgary  \narXiv :2607 .09630v1 [ cs .CV] 10 Jul 2026  \nAbstract  \nCollecting 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 1 . To systematically quantify both factors, we train ResNet-18 modelson a canola branch-counting task using a calibrated L-system plant model 2 . 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. ≈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-correctsand hurts performance.  \nKeywords: L-system, synthetic data augmentation, plant phenotyping, label distribution, ResNet  \n1 Introduction  \nAutomated plant phenotyping via supervised deep learning often requires large annotated datasets, which are costly to produce 3 . Procedural L-system plant models can generate unlimited synthetic images with exact organ-count labels 1;2 , and have been used in ML model training for plant counting tasks in Arabidopsis 1 ,  \ncanola 4 , and wheat 5 . However, due to the domain gap between synthetic and real images, how synthetic data is incorporated has a substantial impact on model performance.  \nPrevious work has identified two particularly important factors for dataset design 1;4 : (i) the ratio of synthetic to real images and (ii) the label distribution of the synthetic dataset. Our work builds on initial explorations by Khan et al. 4 , who used the same canola dataset to investigate how the amount of real images affects performance. By contrast, we focus on synthetic data design and perform a quantitative characterization of the impact of synthetic-to-real ratio and label distribution over a wider range of conditions than previous studies. We identify a broad range of ratios where the addition of synthetic data improves performance. With respect to distributions, we examine the impact of shifting the synthetic label distribution from uniform toward the real data distribution, as well as the effect of Gaussian smoothing applied to the real distribution. Notably, we find that shifting the synthetic distribution away from the actual, observed distribution by smoothing or slightly increasing the representation of underrepresented labels improves model performance.  \n2 Materials and Methods  \nDataset. A dataset of 788 real canola images with ground-truth branch counts (range 1–40) was collected in a high-throughput indoor phenotyping facility 6 . Synthetic images were generated by a calibrated L-system canola model 2 ; Khan et al. 4 calibrated the model parameters to qualitatively match the synthetic branch counts to the real ones. Each plant is rendered at a randomly selected growth day (day 27–50), and the resulting variation in branch count across  \ngrowth stages defines the synthetic distribution used in the ratio sweep experiment. Figure 1 shows the branch-count distributions of both datasets.  \nFigure 1: Branch-count distributions of the real (blue) and synthetic (orange) datasets. The synthetic distribution arises from the L-system model calibrated by Khan et al. 4 .  \nModel. A ResNe","cbCaioY4NTS9xPYt","https://ap.wps.com/l/cbCaioY4NTS9xPYt","pdf",527607,1,5,"English","en",105,"# Abstract\n# Introduction\n# Materials and Methods\n## Dataset\n## Model\n## Exp. 1—Ratio sweep\n## Exp. 2—Distribution interpolation\n## Exp. 3—Gaussian smoothing\n## Exp. 4—Minimum per label","[{\"question\":\"Why do synthetic data and label distribution matter for canola branch counting?\",\"answer\":\"Synthetic images can improve supervised training, but performance depends on how many synthetic samples are used and how the synthetic label frequencies are distributed. Domain gaps make dataset design choices critical.\"},{\"question\":\"What synthetic-to-real ratio yields the best improvement?\",\"answer\":\"The best ratio is 1:7, which reduces mean absolute difference by 7.6% compared with real-only training.\"},{\"question\":\"Which synthetic label distribution strategy performs best?\",\"answer\":\"Gaussian smoothing of the real label distribution yields the best overall result, with abs. diff. of 0.912, improving by 14.7% over real-only training.\"}]",1784180354,13,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"the-effects-of-synthetic-data-and-label-distribution-on-canola-branch-counting","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/the-effects-of-synthetic-data-and-label-distribution-on-canola-branch-counting/82432/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why do synthetic data and label distribution matter for canola branch counting?","Question",{"text":75,"@type":76},"Synthetic images can improve supervised training, but performance depends on how many synthetic samples are used and how the synthetic label frequencies are distributed. Domain gaps make dataset design choices critical.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What synthetic-to-real ratio yields the best improvement?",{"text":80,"@type":76},"The best ratio is 1:7, which reduces mean absolute difference by 7.6% compared with real-only training.",{"name":82,"@type":73,"acceptedAnswer":83},"Which synthetic label distribution strategy performs best?",{"text":84,"@type":76},"Gaussian smoothing of the real label distribution yields the best overall result, with abs. diff. of 0.912, improving by 14.7% over real-only training.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,109,114,119,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":21,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":21,"slug":137},19,"General","general"]