NaturalHistoryMuseum / ALICE

Angled Label Image Capture and Extraction for high throughput insect specimen digitisation
5 stars 1 forks source link

ALICE Software

Installation

pip install -r requirements

Installation errors

Pytorhc is not installed.

Make sure wheel is installed and pip >= 23.2.1 before running pip install -r requirements

Code to extract labels from pinned specimen images.

Aims:

  1. Find boundaries of labels with Mask R-CNN.
  2. Approximate corners of labels.
  3. Transform labels to a 2d viewpoint.
  4. Align labels to a template.
  5. Merge labels.

Converting mask-rcnn region json files into coco/detectron JSON.

python scripts/convert.py data/label

The script will loop through child directories in data/label (data/label/var and data/label/train), read the via_region_data.json, and then output two files coco.json and detectron.json.

Train uses coco.json. Detectron.json is legacy code, and will be removed in future versions.

via_region_data.json must be within the same directory as the images - e.g. data/label/var/via_region_data.json

ALICE Module

Imports all use the alice module namespace - to activate for development:

python setup.py develop

https://github.com/pippy360/transformationInvariantImageSearch/tree/master