Coordinates, image matching, and quality control
coordinates-and-qc.RmdCoordinate errors can leave percent counts numerically unchanged while making the associated image evidence and ML crops wrong. Treat visual QC as a required handoff between import and interpretation.
Two coordinate spaces
Imported .cpc rows can contain:
-
cpce_x,cpce_y: original CPCe display/region coordinates; -
cpce_width,cpce_height: the coordinate-space dimensions used for conversion; -
x_px,y_px: integer coordinates in the matched image; and -
image_width,image_height: image dimensions in pixels.
convert_cpce_coords() uses proportional scaling
independently for each axis.
example_points <- data.frame(
cpce_x = c(0, 25, 50, 100),
cpce_y = c(0, 10, 25, 50)
)
convert_cpce_coords(
example_points,
cpce_width = 100,
cpce_height = 50,
image_width = 1000,
image_height = 500
)
#> # A tibble: 4 × 8
#> cpce_x cpce_y cpce_width cpce_height image_width image_height x_px y_px
#> <dbl> <dbl> <dbl> <dbl> <int> <int> <int> <int>
#> 1 0 0 100 50 1000 500 0 0
#> 2 25 10 100 50 1000 500 250 100
#> 3 50 25 100 50 1000 500 500 250
#> 4 100 50 100 50 1000 500 1000 500The function does not rotate, mirror, crop, georeference, or correct
lens distortion. The current .cpc parser uses ROI maxima as
the coordinate-space dimensions when available because that geometry
fits the bundled fixtures. Other CPCe configurations require
verification.
Basename image matching
match_images() searches under image_root
and matches image_file by basename. This helps when a
.cpc file contains an old Windows path but the same image
now lives on macOS or Linux.
Basenames must be unique. If two directories contain
transect_01.jpg, the first match is used with a warning.
Rename files or separate projects before a release analysis; a
successful basename match does not prove image identity.
Table validation
example_dir <- system.file("extdata", package = "pointcoral")
points <- read_cpce_folder(example_dir, image_root = example_dir, recursive = FALSE)
validate_points(points)
#> # A tibble: 1 × 4
#> check severity n details
#> <chr> <chr> <int> <chr>
#> 1 ok info 0 No validation issues detectedTo see the report structure, introduce deterministic problems in a copy:
bad <- points
bad$raw_label[1] <- NA_character_
bad$x_px[2] <- bad$image_width[2] + 50L
validate_points(bad)
#> # A tibble: 2 × 4
#> check severity n details
#> <chr> <chr> <int> <chr>
#> 1 missing_raw_labels warning 1 Rows with missing raw_label values
#> 2 coordinates_outside_image warning 1 Pixel coordinates outside image boun…Validation reports issues; it does not change the input or decide whether a row should be excluded.
Visual overlay
write_qc_overlays(
points,
image_root = example_dir,
out_dir = "project_outputs/qc",
label_col = "raw_label"
)Review at least:
- points near all four image edges;
- recognizable structures and the expected image orientation;
- labels around mixed or ambiguous benthic cover;
- out-of-frame, repeated, or unexpectedly clustered points; and
- the exact source image and CPCe file identity.
An overlay is evidence that points appear in plausible locations. It cannot confirm sampling randomness, taxonomic correctness, or ecological representativeness.