speciesnet

The speciesnet package provides an R wrapper for Google’s SpeciesNet camera trap AI model, allowing users to classify species in camera trap images using deep learning from within R.

SpeciesNet provides species-level classification (not just animal/person/vehicle detection) and includes the following key features:

• Ensemble approach combining detector + classifier
• Geofencing support to filter predictions by geographic location
• High accuracy trained on millions of camera trap images

Installation

You can install the package from GitHub:

# install.packages("remotes")
remotes::install_github("boettiger-lab/speciesnet")

Setup

The package requires a Python environment. It will automatically install the necessary speciesnet python package:

library(speciesnet)

# Installs speciesnet python package if not present
# Note: This could download ~2GB of dependencies including PyTorch
install_speciesnet()

Usage

library(speciesnet)

# 1. Load the SpeciesNet Model
# Downloads model weights (~214MB) on first use
model <- load_speciesnet()

# 2. Predict species in images
# 2. Predict species in images
# Use an example image included with the package
image_file <- system.file("img", "Caltech_Animal.jpg", package = "speciesnet")
predictions <- predict_species(model, image_file)

# 3. Extract results
# Convert predictions list to a tidy data frame
results_df <- predictions_to_df(predictions)

# View top predictions and detections
print(head(results_df))

Geographic Filtering

You can improve prediction accuracy by providing location information:

# Using country code (ISO 3166-1 alpha-3)
predictions <- predict_species(
  model, 
  image_file,
  country = "USA",
  admin1_region = "CA"  # US state code
)

# Or using coordinates
predictions <- predict_species(
  model, 
  image_file,
  latitude = 38.5,
  longitude = -120.5
)

Batch Processing

Process multiple images at once:

image_files <- c(
  "image1.jpg",
  "image2.jpg", 
  "image3.jpg"
)

predictions <- predict_species(model, image_files)

# Extract results for each image
for (i in seq_along(predictions$predictions)) {
  pred <- predictions$predictions[[i]]
  species <- get_top_species(pred)
  cat(sprintf("Image %d: %s\n", i, species))
}

About SpeciesNet

SpeciesNet is developed by Google Research and provides state-of-the-art species classification for camera trap images. The model:

• Uses SpeciesNet v4.0.2a by default (always-crop ensemble)
• Supports 1000+ species across multiple continents
• Provides detection bounding boxes and species classifications
• Can be geofenced to improve accuracy in specific regions

For more information, see the official SpeciesNet repository.

License

MIT