Real-time detection from camera feeds and live sources
Good accuracy with custom-trained models. Handles segmentation and rotated boxes. Strong across sizes, especially large objects.
Best for: accuracy and aerial / specialized models.
Capacity: ~1 camera / GPU
Measured on a Tesla T4 (AWS g4dn) at 1080p with tracking. Varies with your hardware and the model size you pick.
Select a model above to filter classes
No Streams Active
Configure a source on the left and click Add Stream