This zipfile contains videos to support the UAI2023 submission with the title:
Heteroskedastic Geospatial Tracking with Distributed Camera Networks

The videos are split into the two data scenario: lights_off and lights_on

In each directory there is a directory for each model: ResNet50, DETR, and Ours

In each of thoese directories there is two videos: det_vid.mp4 and kf_vid.mp4

det_vid.mp4 shows the predicted distributions from the 4 independent detector models (in black)

kf_vid.mp4 shows the predicted distribution from the Kalman Filter (in red)

The green box denotes the physical extent of the object with an arrow indicting the object's heading

All distributions are calibrated using NLL

The plotted ellipse displays the 95% confidence region of the distribution

Each video shows the first 500 frames of the test set

Due to space constraints there is non-trivial compression applied
