Datasets:
Need Help: 3D HOTA Evaluation Script for Validation Split
Hi everyone,
I'm currently working on evaluating my algorithm on the validation set, but I'm unsure how to correctly compute performance metrics—especially for 3D HOTA in a multi-camera setup. I've reviewed the evaluation instructions but couldn’t find a clear example or script that handles this case.
Could anyone kindly clarify:
How is 3D HOTA computed in multi-camera settings?
Is there an official or recommended script available to evaluate the algorithm on the validation split?
A sample evaluation script or pipeline would be greatly appreciated. Thanks in advance!
The official evaluation script for the validation set will be included in an upcoming release.
In the meantime, you can refer to the HOTA evaluation toolkit as a starting point. For the multi-camera 3D setting, the key modification is to replace the 2D bounding box IoU computation with 3D bounding box IoU. Each scene can be treated as a single camera in the global coordinate system during evaluation.
Hi,
While the evaluation script for this year's dataset is to be released later, I'd like to ask if last year's 2024 evaluation script (one that gauges the ID based on the target's ground-plane location) can be released soon as well. As far as I know, it was released to last year's participants during the 2024 challenge.
Hi,
Till when will the script for 3D HOTA metric be released ?
@xifen527
@VDotMint
@agntgalahad
We have pushed the evaluation scripts for the 2024 edition. The 2025 version will also be updated shortly.