Barbados Traffic Analysis Challenge
'/%3e%3c/svg%3e)
Helping Barbados
$11 000 USD
Completed (4 months ago)
Computer Vision
Prediction
Object Tracking
Video Analytics
Deep Learning
1840 joined
216 active
Start
Nov 07, 25
Close
Jan 25, 26
Reveal
Jan 26, 26
Defining ROIs manually
Data · 24 Nov 2025, 13:03 · 3
Is it allowed to define ROIs manually? I know strictly speaking this would fall under "manual labeling" which is not allowed. I still want to make sure that this really means any labeling and not only on a video level or frame level for example for annotating cars.
Another question:
Am I allowed to create global heatmaps of the object positions? Like let's say I use some detection model and then over all videos for a specific camera I create a global heatmap of the positions for the detected labels. That would probably not be allowed because if I use that for inference I would use data from future videos, right?
Hi @cytox
Thanks for the query!
Defining ROIs manually does count as manual labeling under the rules. Since all data manipulation must be done in code, ROIs need to be defined programmatically.
The restriction applies to any human‑drawn or Excel‑based manipulation. To stay compliant, you should generate ROIs through code, not by hand.
You can safely use global heatmaps since they are generated entirely from the training & test input videos only. Since you do not have access to any test-segment frames (minutes 18–22), there is no possibility of incorporating future information, so using a training-derived heatmap during inference is allowed.
I think visualizing the traffic patterns before jumping into modeling is really important here. EDA can help uncover things like peak congestion times or unusual flow behaviors that a simple metric might miss. Also, it might be worth trying time-based features (e.g., hour of day, day of week) and spatial context (e.g., nearby junction types) in your model to capture patterns better. Looking forward to seeing what others have found PolyTrack
@paper minecraft I believe it's crucial to visualize the traffic patterns before diving into modeling.