cam2model
Preprocessing Trap Camera Data for Machine Learning Models
Provides tools for preprocessing camera trap image data to prepare it for training machine learning models. It automates tasks such as extracting EXIF metadata, processing and analyzing image attributes, selecting relevant images (e.g., daytime photos), sampling subsets, and managing file organization. These features streamline the workflow for ecological research and wildlife monitoring projects. Additionally, the package includes tools for generating exploratory visualizations and creating annotation datasets.
cam2model
cam2model provides a workflow for preparing camera trap images for annotation and model training.
Recommended workflow
rename_images()process_metadata()generate_plot()(optional)process_training_images()copy_in_batches()
Quick reference by stage
| Stage | Minimum input | Main output | File operations |
|---|---|---|---|
rename_images() |
Directory containing images | Result tibble (or a list with plan and result) |
Creates a site folder in output_dir and copies images |
process_metadata() |
Directory of renamed images | Metadata tibble | Writes the output_file CSV |
generate_plot() |
Data frame with the columns used for aesthetics | ggplot object |
No files written |
process_training_images() |
CSV with Directory and FileName |
Sample tibble | Creates dest_dir, writes two CSV files, and optionally copies images |
copy_in_batches() |
Data frame with id and path |
Per-file report data frame | Creates batch subfolders and copies files (or simulates the operation with dry_run) |
End-to-end example (temporary paths)
library(cam2model)
# 1) Set up temporary paths
raw_dir <- tempfile("cam2model_raw_")
renamed_dir <- tempfile("cam2model_renamed_")
train_dir <- tempfile("cam2model_train_")
batch_dir <- tempfile("cam2model_batches_")
dir.create(raw_dir, recursive = TRUE)
dir.create(renamed_dir, recursive = TRUE)
dir.create(train_dir, recursive = TRUE)
# Copy real camera trap images into raw_dir before running the next stages
# file.copy("path_to_real_image.jpg", file.path(raw_dir, "cam1", "image.jpg"), recursive = TRUE)
# 2) Rename and flatten (requires real images)
# renamed <- rename_images(input_dir = raw_dir, output_dir = renamed_dir, site_name = "demo")
# 3) Extract metadata (requires real EXIF data)
# metadata <- process_metadata(output_dir = renamed_dir, output_file = "metadata.csv", UserComment = FALSE)
# metadata_csv <- file.path(renamed_dir, "metadata.csv")
# 4) Self-contained training example without real EXIF data
src_dir <- tempfile("cam2model_src_")
dir.create(src_dir, recursive = TRUE)
file_a <- file.path(src_dir, "a.jpg")
file_b <- file.path(src_dir, "b.jpg")
writeBin(charToRaw("a"), file_a)
writeBin(charToRaw("b"), file_b)
metadata_tbl <- data.frame(
Directory = c(src_dir, src_dir),
FileName = c("a.jpg", "b.jpg"),
File_hour = c(8, 11),
stringsAsFactors = FALSE
)
metadata_csv <- tempfile("cam2model_metadata_", fileext = ".csv")
readr::write_csv(metadata_tbl, metadata_csv)
sampled <- process_training_images(
metadata_file = metadata_csv,
dest_dir = train_dir,
output_sample_file = "sampled_images.csv",
output_annotation_file = "annotation_data.csv",
sample_proportion = 1,
copy_images = FALSE,
show_progress = FALSE
)
# 5) Reorganize into batches (dry run)
batch_report <- copy_in_batches(
df = data.frame(
id = sprintf("img_%03d", seq_len(nrow(sampled))),
path = sampled$file_path,
stringsAsFactors = FALSE
),
output_dir = batch_dir,
batch_size = 250L,
dry_run = TRUE,
verbose = FALSE
)
head(batch_report)
Workflow vignette
See vignettes/cam2model-workflow.Rmd for a step-by-step guide with a table of inputs, outputs, and generated files.