Hi all, we have gone through the AutoScenes dataset and met problems below.
We have also start an issue at https://github.com/SSLAD2021/AutoScenes-eval/issues/1 to paste the images.
### Q1: Unified Model
Is it permitted to utilize individual model for object detection and road segmentation?
### Q2: Inaccurate Box annotations
We found that some inaccurate 3D bounding boxes when projecting them to images. For example **scene-plus-0751**. Generally inaccurate extrinsic or asynchronous sensor timestamps my cause the problem. Is that negligible for training?
https://user-images.githubusercontent.com/30251143/182871279-a7b9a1c4-0b0b-4d94-b640-0c35ceb3cdf4.png
And there are also some strange rotating boxes. For example **scene-plus-0734**.
https://user-images.githubusercontent.com/30251143/182875773-138769cc-3b37-42b2-a8db-4f414ce71716.png
### Q3: Invalid Segmentation Polygons
Following the instructions(https://codalab.lisn.upsaclay.fr/competitions/6674#participate), we use nuScenes devkit to extract the bev segmentation mask. However, we found that only 10 scenes (scene-plus-0670, scene-plus-0676, scene-plus-0696, scene-plus-0705, scene-plus-0729, scene-plus-0742, scene-plus-0751, scene-plus-0826, scene-plus-0844, scene-plus-0874) out of 45 trainval scenes have the correct bev mask as paste below. Somehow we extract **blank mask** for all segmentation classes from the rest of the trainval scenes.
https://user-images.githubusercontent.com/30251143/182877183-df3df2fa-fbb6-4e16-a678-5a5cab1c29f8.png
### Q4: Missing Ped-Crossing
It seems that no "ped-crossing" is annotated in trainval scenes, which conflicts with what challenge claims.
### Q5: Unknown Keywords
We have not figured out the meaning of "relative translation" and "relative rotation" in ego_pose.json and "relative_velocity" in sample_annotation.json
Please follow the issue https://github.com/SSLAD2021/AutoScenes-eval/issues/1
Posted by: SSLAD2022 @ Aug. 7, 2022, 6:25 a.m.