The charts explained

A project turns your results into charts. The Dashboard gives you the overview, and the Insights pages go deeper, each with its own filters.

The screenshots below come from a demo survey of 12 camera sites across Pennsylvania. The photos are real, from the ENA24 dataset, but we edited the capture times in the files to build a survey that spreads over a season and several cameras. So the animals and the species are genuine, while the dates, trends and trap nights are made up for the demo.

Dashboard

The Dashboard has two tabs.

Overview shows the whole project, without filters. You see a random animal photo (a new one every time you open the dashboard, or press the refresh button for another), how many files were recorded and how many trap nights that took, a calendar of when your cameras were running (one row per year, one square per week, darker for more cameras), the top ten wildlife taxa, and how much of the work you have verified. Pick a higher rank in the chart to compare groups, such as mammals against birds at class level. Click a bar to explore it.

Explore shows every label by default, people and vehicles included, at the sites, site tags and dates you choose. Pick labels in the same tree as on the Counts page to narrow it: one species, a whole family, a set like livestock, or everything except people and vehicles. Every card follows the same choice: the number of events and observations, each per 100 trap nights, random photos, the day and night activity pattern, a trend over time, and a small map of where it was seen.

The photos are picked at random from detections the AI is sure about, so the name under a photo is one the AI is confident of. Click a photo to open the file on the Labels page and check it.

Explore also shows sex, age and behaviour. The AI does not fill these in, so they read Unknown until you set them per event on the Counts page.

The Overview's chart and random animal show wildlife only, which their titles say. Labels like blank or false detection never count anywhere. A class that only comes from the detector, like "shark" from a marine detector without a classifier, counts as wildlife but cannot be picked on its own yet.

Map

Each camera site on a map, coloured by its observation rate per 100 trap nights. You can switch how the sites are drawn: one marker per site, nearby sites merged into a hex grid, or nearby sites grouped into clusters. Deployments with no site do not appear. Map style changes the background: Light, Satellite, or OpenStreetMap. Satellite is the useful one when you want to see the habitat around a camera, like where the forest gives way to open ground.

Activity overlap

Compare the daily activity of one or two species. With two species selected, the chart also shows the overlap coefficient: 0 means the two are never active at the same hours, 1 means their patterns are identical. Each species is also labelled diurnal (day), nocturnal (night), crepuscular (dawn and dusk) or cathemeral (no clear preference, active day and night). In the demo, the opossum comes out nocturnal and the chipmunk diurnal.

You can read the day on two different clocks. Clock time is the plain 24 hour day. Sun time stretches each day so that sunrise and sunset line up across the whole survey. Use it when your data spans seasons or sites: 06:00 in December and 06:00 in June are not the same moment relative to dawn, and clock time would blur that difference away. It needs site coordinates to work out the sun.

Deployment timeline

One bar per deployment, showing when each camera was active and what it recorded. Below it you see how many cameras ran at the same time, which is the quickest way to spot a gap in your survey effort.

Confusion matrix and class performance

Where you have verified labels, these two views compare the AI's labels against your verified ones. The confusion matrix shows where the AI mixes up species. Class performance shows accuracy per species. Both only use detections you verified, so they stay empty until you have checked some labels.

Insights need capture times and, for the map, site coordinates. Files with no readable timestamp are still detected and classified, but they drop out of the time-based views.