Extract Munsell color from a profile photo — extract_munsell_from

Sends the photo to a multimodal VLM with a prompt that asks the model to estimate Munsell hue / value / chroma per visible horizon (when a Munsell reference card is in frame). Recorded as extracted_vlm with the model's self-reported confidence; photos without a reference card should yield confidence below 0.5 per the prompt specification.

Usage

extract_munsell_from_photo(
  pedon,
  image_path,
  provider,
  max_retries = 3L,
  overwrite = FALSE,
  prompt_name = "extract_munsell_from_photo",
  schema_name = "munsell"
)

Arguments

pedon: A PedonRecord.
image_path: Path to the image file (JPG / PNG).
provider: A chat provider from vlm_provider (or a MockVLMProvider for testing).
max_retries: Integer; how many times to re-prompt on validation failure. Default 3.
overwrite: If TRUE, lower-authority values are allowed to clobber higher-authority ones. Default FALSE.
prompt_name: Override the default prompt template ("extract_horizons").
schema_name: Override the default schema ("munsell", a colour-only subset of the horizon schema: geometry, designation and Munsell only). Since v0.9.194 – the full horizon schema was injected into the prompt and all but those fields discarded, which on a token-metered endpoint cost roughly 2,500 tokens per call for attributes a photograph cannot show.

Value

Invisibly, the mutated pedon, with the photo added to pedon$images.

Details

Quantitative non-color attributes (clay %, CEC, pH, etc.) are never extracted from photos, by prompt-level instruction. If the model returns one anyway, it is silently dropped.