FishMorph

Import images, mark landmarks and body outlines, then save measurements with the specimen record. FishMorph retains the original image and keeps a history of annotation changes.

The bundled demo uses synthetic fish, so it is easy to try without supplying photographs.

Landmarks and a reviewed body outline give length-relative measurements and profiles. Multiple photographs of one specimen stay grouped together rather than being counted as separate animals.

Measurements depend on the available landmarks and scale. An unscaled photograph cannot give physical length or reveal the width of an unseen side.

python3 -m venv .venv
.venv/bin/python -m pip install -e ".[dev]"
.venv/bin/fishmorph --project ./demo-data serve

Load a mesh to render orthographic views on the CPU, then mark matching landmarks and compare its proportions with a reference profile. The saved projection matrices make each view reproducible.

Color comparison needs calibrated reference data. A neutral model render and an ordinary field photograph usually cannot provide that comparison.

The packaged application passed 68 tests. Its demo contains eight reference specimens, nine images and a model comparison.

Browser checks covered annotation saving, measurements, zoom, masks and JSON export at desktop and phone widths. Exported JSON matched the analysis stored in SQLite.

.venv/bin/python -m pytest -q

The code and bundled synthetic examples are MIT licensed. Imported photographs and meshes keep their original licenses.

Manual landmarks and masks need review. The examples test the software workflow; they do not establish biological accuracy.