Loading latest highlights...
For each bird species we detect, we find the "best sighting" by analysing all the moments where that bird appeared in the video.
How it works:
We look at every possible 10-15 second window in the video and add up the confidence scores for each detection of that species within the window. The window with the highest total score becomes the "best sighting" for that bird.
This means the best moment is usually when the species appears multiple times, stays in view longer, or when we're very confident in the identification - or a combination of all three.
When you click on a species name, the video jumps to that moment so the following few seconds are the best view this highlights video can offer of that type of bird.
The same species may also occur in other parts of the highlights video.
Very Probable detections are species that appeared frequently in the highlights video.
These species were detected in 4 or more separate moments, indicating they were present around the feeder multiple times during the recording period.
The frequency of detections makes these identifications highly reliable, as the bird was captured from multiple angles or positions.
Probable detections are species identified with high confidence (80% or above) but appeared fewer than 4 times in the video.
While we have strong confidence in the identification based on visual features, these birds were seen less frequently than "Very Probable" detections.
This typically indicates a brief visit or a bird that stayed off-camera for most of its time near the feeder.
Possible detections are species identified with moderate confidence (below 80%) and appeared fewer than 4 times.
These identifications are less certain due to factors like:
While these detections may be accurate, they should be viewed with some scepticism and ideally verified by reviewing the video footage.
Bird species are visually identified using BioCLIP analysis of motion-triggered video clips.
Detections represent birds visible on camera during recording periods.
The confidence percentages indicate how certain the AI model is about each identification, based on visual features like plumage patterns, body shape, and size.
Bird vocalisations are detected from background audio in motion-triggered video clips using BirdNET acoustic analysis.
How it works:
Audio is extracted from each video clip and analysed for bird songs, calls, and other vocalisations. The AI model identifies species based on acoustic patterns including frequency, rhythm, and tonal characteristics.
Important context:
These detections represent birds that were vocalising nearby during the recording, but were likely off-camera. The motion trigger that started the video was typically activated by a different bird or movement at the feeder.
Audio detections complement the visual detections by revealing the presence of species that may not have approached the feeder, but were active in the surrounding area.
The confidence percentages indicate how certain the AI model is about each species identification based on the acoustic signature of the vocalisation.