Image & Video Annotation

Create and curate labelled observations on both still imagery and video frames — the optical "ground truth" that anchors the acoustic interpretation.

Image annotation

Three detector boxes on one frame, with the editor alongside. The Annotated Images list jumps straight to any annotated frame; the panel below it lists this frame's labels and scores — fish (0.86), fish (0.75), fish (0.61). Note the last one against the Detection Confidence of 0.60: raise the threshold by a hundredth and that box disappears.

• Draw a bounding box over a feature in the current image, assign a label and a confidence, and save it.
• Existing annotations (e.g. from an automated detector CSV) are shown as boxes; select one to edit its label/extent or delete it.
• A confidence threshold filters which annotations are displayed (shared with the Image Browser).
• Known labels are remembered to keep labelling consistent across images.

How to use: enter Add (draw) mode, drag a box on the image, pick/enter a label and confidence in the editor panel, and save. Click an existing box to re-select it for editing or deletion.

Video annotation

Per-frame annotations in the Video Player.

• Annotations are attached per frame: add / edit / delete boxes on the frame currently shown in the Video Player.
• Each annotation carries a label and confidence, mirroring the image workflow.

Annotation data

Image annotations come from / persist to the imageannotation CSV; video annotations are keyed by frame index. Both feed the same labelled-observation model used elsewhere in the plugin.

The two file formats are different and are documented column by column in Annotations & video — the image CSV is read positionally (column order matters, header names do not), the video CSV by header name.