search

tnc-ca-geo / animl-frontend

A frontend web app for viewing & labeling camera trap data by The Nature Conservancy.
https://animl.camera
14 stars 0 forks source link

Support export to Camtrap DP #199

Open abfleishman opened 4 months ago

abfleishman commented 4 months ago

Camtrap DP defines an open frictionless camera trap data exchange format. It would be really nice if animl could export data to this format easily (as well as reading data from this format!).

See this repo

nathanielrindlaub commented 4 months ago

@abfleishman interesting - that spec is new to me but I'm taking a look now. At first glance this should be possible.

This is somewhat subjective but I don't love that it seems like the lowest level the observationLevel field supports is to the media level (i.e., you can associate an observation with an image). For ML training purposes it's much more useful to go down to the object level (i.e., associate an observation/label with a localized bounding box within an image), and Animl's data model natively supports this, as does the COCO for camera traps JSON format that you can export from Animl.

In my (admittedly somewhat limited - I'm not an ML engineer) experience the best practice is to crop out animals from their backgrounds when training classifiers, so having that level of granularity as well as bounding box coordinates in your label/annotation data helps a ton.

abfleishman commented 3 months ago

@nathanielrindlaub They do support object (bounding box) level annotations See the bbox section and down