01
Set up analysis
Choose footage and sampling settings.
Works best with a fixed camera. Draw a crop around the tank floor. Fewer than five blobs may mean fish overlap; extra blobs may come from tank edges or reflections.
Model settings Roboflow InferenceJS
Using zebrafish_in_aquarium/9, a trained YOLOv8n model for zebrafish. Enter a Roboflow workspace publishable key, never a private API key.
Add a video to begin.
02
Frame preview
Detected fish centers and nearby links.
◎Waiting for footageYour analyzed frames will appear here.
Drag on the preview to select the tank area.
—
ANALYSIS COMPLETE
What the frames show
How metrics work
Each detection center is a candidate fish position. Pair distance is the mean of all detection pairs within a frame, then averaged over frames with at least two detections. “Near another” is the share whose nearest neighbor is within the chosen distance. Connected detections form a cluster; single detections are excluded. Inch estimates use the paper-grid calibration and are approximate. When the detected count is low, the detector checks close to the crop edge, then may split a large joined region. These estimates are marked in the raw export. Reflections can create extras, and fish are not tracked between frames.