Wrong Visualization on Training Set

mit-han-lab / bevfusion

[ICRA'23] BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's-Eye View Representation

https://bevfusion.mit.edu

Apache License 2.0

2.37k stars 427 forks source link

Closed NoOneUST closed 6 months ago

NoOneUST commented 1 year ago

I can generate correct visualizations on validation and testing sets, but the results are wrong on training sets.

1526915243047392-f9878012c3f6412184c294c13ba4bac3 1526915251397102-366efab9d8e4475e9a5f28c7ae7850f7

The code I use is the stable branch.

These two images are the gt and pred results on the training set. I have confirmed that this error impacts all images in the training set.

Anna1ximi commented 1 year ago

@NoOneUST Hey! I want to know how you modified visualize.py for visualization. I have been stuck on this question for days. Thanks!

san9569 commented 1 year ago

I'm not author or contributor of this project.

In my opinion, it seems to be related to data augmentation. Training data is processed by specific pipeline which contains augmentation.