deploy_model: Deploy model on camera trap images

In TabakM/CameraTrapDetectoR: CameraTrapDetectoR

Deploy model on camera trap images

Description

This function deploys a model trained to identify and count the objects in camera trap images.

Usage

deploy_model(
  data_dir = NULL,
  model_type = "general",
  recursive = TRUE,
  file_extensions = c(".jpg", ".JPG"),
  make_plots = TRUE,
  plot_label = TRUE,
  output_dir = NULL,
  sample50 = FALSE,
  write_bbox_csv = FALSE,
  overlap_correction = TRUE,
  overlap_threshold = 0.9,
  score_threshold = 0.6,
  return_data_frame = TRUE,
  prediction_format = "wide",
  h = 307,
  w = 408,
  lty = 1,
  lwd = 2,
  col = "red",
  labeled = FALSE
)

Arguments

data_dir	Absolute path to the folder containing your images
model_type	Options are c('general', 'species', 'family'). The 'general' model predicts to the level of mammal, bird, humans, vehicles. The 'species' model recognizes 77 species. The 'family' model recognizes 33 families.
recursive	boolean. Do you have images in subfolders within your data_dir that you want to analyze, if so, set to TRUE. If you only want to analyze images within your data_dir and not within sub-folders, set to FALSE.
file_extensions	The types of extensions on your image files. Default is c(".jpg", ".JPG")
make_plots	boolean. Do you want to make plots of the images with their predicted bounding boxes?
plot_label	boolean. Do you want the plots to contain the predicted class of object
output_dir	You can specify absolute path to output. Default is 'NULL', and creates a folder within your data_dir. Only specify a path if you want the results stored somewhere else on your computer.
sample50	boolean. Do you want to run the model only on a subset of 50 images from your dataset? This is a good idea if you are experimenting with settings.
write_bbox_csv	boolean. Do you want to create a csv with all of the information on predicted bounding boxes? This csv will include all bounding boxes, even those with low probability.
overlap_correction	boolean. Should overlapping detections be evaluated for overlap and highest confidence detection be returned
overlap_threshold	Overlap threshold used when determining if bounding box detections are to be considered a single detection. Accepts values from 0-1 representing the proportion of bounding box overlap.
score_threshold	Confidence threshold for using a bounding box. Default is 0.6. A lower number will produce more bboxes (it will be less stringent in deciding to make a bbox). A higher number will produce fewer bboxes (it will be more stringent).
return_data_frame	boolean. Do you want a dataframe returned
prediction_format	The format to be used for the prediction file. Accepts values of 'wide' or 'long'.
h	The image height (in pixels) for the annotated plot. Only used if make_plots=TRUE.
w	The image width (in pixels) for the annotated plot.
lty	line type for bbox plot. See ?plot for details
lwd	line width for bbox plot. See ?plot for details
col	line color for bbox plot. See ?plot for details
labeled	This is not functional

Details

This function deploys a model to detect and classify objects in camera trap images. The function will find all files matching the 'file_extension's specified within the 'data_dir' specified and deploy the 'model_type' on these images. If you specify recusive=TRUE, the function will find relevant image files within all subdirectories of your 'data_dir'. 'deploy_model' returns a dataframe of predicted number of individuals within each category in each image. This dataframe is also written as a csv file within your 'output_dir'. If you specify make_plots=TRUE, the function will plot predicted bounding boxes for each image in your 'output_dir'. If you are working with many images, you may wish to specify sample50=TRUE the first time you use this function, which will only deploy the model on 50 of your images. There are three options for model_type: 'general' recognizes mammals, birds, humans, and vehicles. 'species' recognizes 77 species. 'family' recognizes 33 families. If you want to see all of the information for each bounding box (including coordinates, labels, and confidence), specify write_bbox_csv=TRUE and it will be produced in your 'output_dir'. Additionally, A file called "arguments" will be produced in your 'output_dir' this is a list of all of the arguments you passed to this function for reference.

Value

Returns a dataframe of predictions for each file. The rows in this dataframe are the file names in your 'data_dir'; the columns are the categories in the model. If any of your images were not loaded properly, there will be a column in the dataframe called 'image_error'. Images with a 1 in this column had issues and the model was not deployed on them.

TabakM/CameraTrapDetectoR documentation built on June 11, 2022, 9:37 p.m.