pliman

The pliman package offers tools for both single and batch image manipulation and analysis (Olivoto, 2022), including the quantification of leaf area, disease severity assessment (Olivoto et al., 2022), object counting, extraction of image indexes, shape measurement, object landmark identification, and Elliptical Fourier Analysis of object outlines, as detailed by Claude (2008). The package also provides a comprehensive pipeline for generating shapefiles with complex layouts and supports high-throughput phenotyping of RGB, multispectral, and hyperspectral orthomosaics. This functionality facilitates field phenotyping using UAV- or satellite-based imagery.

pliman also provides useful functions for image transformation, binarization, segmentation, and resolution. Please visit the Examples page on the pliman website for detailed documentation of each function.

Installation

Install the latest stable version of pliman from CRAN with:

install.packages("pliman")

The development version of pliman can be installed from GitHub using the pak package:

# pak::pkg_install("nepem-ufsc/pliman")
install.packages("pliman", repos = c('https://nepem-ufsc.r-universe.dev', 'https://cloud.r-project.org'))

Note: If you are a Windows user, you should also first download and install the latest version of Rtools.

Analyze objects

The function analyze_objects() can be used to analyze objects such as leaves, grains, pods, and pollen in an image. By default, all measures are returned in pixel units. Users can adjust the object measures with get_measures() provided that the image resolution (Dots Per Inch) is known. Another option is to use a reference object in the image. In this last case, the argument reference must be set to TRUE. There are two options to identify the reference object:

By its color, using the arguments back_fore_index and fore_ref_index
By its size, using the arguments reference_larger or reference_smaller

In both cases, the reference_area must be declared. Let’s see how to analyze an image with flax grains containing a reference object (rectangle with 2x3 cm). Here, we’ll identify the reference object by its size; so, the final results in this case will be in metric units (cm).

library(pliman)
img <- image_pliman("flax_grains.jpg")
flax <- 
  analyze_objects(img,
                  index = "GRAY",
                  reference = TRUE,
                  reference_larger = TRUE,
                  reference_area = 6,
                  marker = "point",
                  marker_size = 0.5,
                  marker_col = "red", # default is white
                  show_contour = FALSE) # default is TRUE

# summary statistics
flax$statistics
#        stat        value
# 1         n 2.680000e+02
# 2  min_area 3.606989e-02
# 3 mean_area 6.250403e-02
# 4  max_area 1.262446e-01
# 5   sd_area 8.047152e-03
# 6  sum_area 1.675108e+01
# 7  coverage 5.388462e-02

Disease severity

Using image indexes

To compute the percentage of symptomatic leaf area you can use the measure_disease() function you can use an image index to segment the entire leaf from the background and then separate the diseased tissue from the healthy tissue. Alternatively, you can provide color palette samples to the measure_disease() function. In this approach, the function fits a general linear model (binomial family) to the RGB values of the image. It then uses the color palette samples to segment the lesions from the healthy leaf.

In the following example, we compute the symptomatic area of a soybean leaf. The proportion of healthy and symptomatic areas is given as a proportion of the total leaf area after segmenting the leaf from the background (blue).

img <- image_pliman("sev_leaf.jpg")
# Computes the symptomatic area
sev <- 
  measure_disease(img = img,
                  index_lb = "B", # to remove the background
                  index_dh = "NGRDI", # to isolate the diseased area
                  threshold = c("Otsu", 0), # You can also use the Otsu algorithm in both indexes (default)
                  plot = TRUE)

sev$severity
#    healthy symptomatic
# 1 92.63213    7.367868

Interactive disease measurements

An alternative approach to measuring disease percentage is available through the measure_disease_iter() function. This function offers an interactive interface that empowers users to manually select sample colors directly from the image. By doing so, it provides a highly customizable analysis method.

One advantage of using measure_disease_iter() is the ability to utilize the “mapview” viewer, which enhances the analysis process by offering zoom-in options. This feature allows users to closely examine specific areas of the image, enabling detailed inspection and accurate disease measurement.

img <- image_pliman("sev_leaf.jpg", plot = TRUE)

# works only in an interactive section
measure_disease_iter(img, viewer = "mapview")

Deep Learning & Foundation Models

pliman integrates pre-trained Computer Vision and Deep Learning Foundation models running directly in C++ via Microsoft ONNX Runtime, with zero Python or PyTorch dependency. Models automatically execute with CPU multi-threading or hardware-accelerated Windows DirectML GPU support (engine = "gpu").

Models are downloaded automatically on first use or manually via pliman_download_model().

Task / Capability	Function	Supported Models	Description
Object Detection	`image_detect_dl()`	`yolo26[n/s/m/l/x]`	Real-time object detection with bounding boxes (80 COCO classes).
Instance Segmentation	`image_segment_dl()`	`yolo26[n/s/m/l/x]-seg`	Real-time instance segmentation with continuous bilinear sub-pixel mask interpolation and 28 morphological features.
Pose Estimation	`image_pose_dl()`	`yolo26[n/s/m/l/x]-pose`	Real-time human pose estimation with 17 COCO keypoints and skeleton visualization.
Image Classification	`image_classify_dl()`	`yolo26[n/s/m/l]-cls`	Fast ImageNet 1,000-class image classification.
Star-Convex Segmentation	`image_stardist_dl()`	`stardist`, `stardist-dsb2018`	48-ray star-convex polygon detection for round, convex, or touching grains, seeds, and cells.
3D Depth Estimation	`image_depth_dl()`	`depth-anything-v2`	Monocular 3D relative depth estimation (Depth Anything V2 Small, 518×518).
Semantic Foundation Features	`image_features_dl()`	`dinov2`	Dense 384-D patch embeddings and top-3 PCA false-color semantic segmentation via DINOv2 (ViT-S/14).
4x Super-Resolution	`image_superres_dl()`	`realesrgan-compact`	4x generative convolutional super-resolution with memory-efficient tiled inference.
Text-Prompted Segmentation	`image_segment_dl()`	`grounded-sam`	Zero-shot open-vocabulary instance segmentation via Grounding DINO + SAM 2.1.
One-Shot Exemplar Counting	`image_segment_dl()`	`persam`	One-shot visual exemplar segmentation and counting via SAM 2.1 and Dense Conv ZNCC.
Salient Background Removal	`image_remove_bg()`	`ben2`, `u2netp`, `rmbg-2.0`	Ultra-sharp boundary cutouts and transparent background export.

1. Real-Time Detection, Segmentation, Pose, and Classification (YOLO26)

library(pliman)

img <- image_import("objects.jpg")

# Fast bounding box detection
boxes <- image_detect_dl(img, model = "yolo26n", conf_threshold = 0.25)

# High-resolution instance segmentation with polygon overlay
res_hl <- image_segment_dl(
  img,
  model = "yolo26n-seg",
  type = "highlight",
  box_threshold = 0.25,
  bbox = TRUE
)

# Single multi-label mask image with integer labels (0 = background, 1..N = instances)
res_mask <- image_segment_dl(img, model = "yolo26n-seg", type = "mask")

# Inspect morphological features (area, perimeter, circularity, etc.)
head(res_hl$features)

# Real-time pose estimation (17 COCO keypoints + skeleton)
pose_res <- image_pose_dl(img, model = "yolo26n-pose", conf_threshold = 0.3)

# 1,000-class ImageNet classification
top_classes <- image_classify_dl(img, model = "yolo26n-cls", top_k = 5)

2. Monocular 3D Depth Estimation (Depth Anything V2)

# Computes relative 3D depth and displays with an inferno colormap
depth_img <- image_depth_dl(img, col_palette = "inferno")

# Access raw continuous depth values
raw_depth <- attr(depth_img, "depth")

3. Foundation Semantic Embeddings (DINOv2)

# Extract 384-dimensional ViT patch embeddings and top-3 PCA false-color map
feats <- image_features_dl(img, return_pca = TRUE)

# Global image classification token vector
feats$cls_token

4. Star-Convex Polygon Segmentation (StarDist)

# Ideal for crowded seeds, grains, or cell nuclei
grains <- image_import("grains.jpg")
res_star <- image_stardist_dl(grains, prob_threshold = 0.5)

5. Generative 4x Super-Resolution (Real-ESRGAN Compact)

# Upscale low-resolution imagery by 4x with tiled memory-safe inference
img_4x <- image_superres_dl(img, scale = 4)

Citation

citation("pliman")
Please, support this project by citing it in your publications!

  Olivoto T (2022). "Lights, camera, pliman! An R package for plant
  image analysis." _Methods in Ecology and Evolution_, *13*(4),
  789-798. doi:10.1111/2041-210X.13803
  <https://doi.org/10.1111/2041-210X.13803>.

Uma entrada BibTeX para usuários(as) de LaTeX é

  @Article{,
    title = {Lights, camera, pliman! An R package for plant image analysis},
    author = {Tiago Olivoto},
    year = {2022},
    journal = {Methods in Ecology and Evolution},
    volume = {13},
    number = {4},
    pages = {789-798},
    doi = {10.1111/2041-210X.13803},
  }

Getting help

If you come across any clear bugs while using the package, please consider filing a minimal reproducible example on github. This will help the developers address the issue promptly.

Suggestions and criticisms aimed at improving the quality and usability of the package are highly encouraged. Your feedback is valuable in making {pliman} even better!

Code of Conduct

Please note that the pliman project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.