This page is the website-facing catalog for the petrographer model and dataset library. It is designed for browsing what is available, checking high-level training details, and copying the right from_pretrained() or train_model() calls into your own workflow.
Available Models
Available Datasets
Loading a Model
All models can be loaded with a single line:
library(petrographer)
# Load any model from the hub
model <- from_pretrained("model_id")
# Run predictions
results <- predict(model, "path/to/image.jpg")Publishing to the Library
Maintainers can publish a finished local model pin to the shared library board with pin_model(). The exact destination board depends on the deployment environment, so the example below is intentionally generic:
# Download the local pin so you can locate the model directory
local_board <- pins::board_folder(here::here(".petrographer/models"), versioned = TRUE)
files <- pins::pin_download(local_board, "my_model")
model_dir <- fs::path_dir(files[[1]])
# Configure the destination board
# board <- pins::board_url("https://example.com/petrographer/models/")
# board <- pins::board_folder("~/shared/petrographer-models", versioned = TRUE)
# Publish the local pin to that board
pin_model(
model_dir = model_dir,
model_id = "my_model",
board = board,
metadata = list(
description = "Brief description",
model_variant = "small"
)
)
# Refresh the destination board manifest for board_url() consumers
pins::write_board_manifest(board)Contact the package maintainers to have your models added to the public hub.