Open GG-Bonds opened 1 month ago
I noticed that 《Visual Point Cloud Forecasting enables Scalable Autonomous Driving》 This article only loads the pre-trained 2D backbone, and the effect of adding points is very weak. Are the authors aware of similar ablation experiments performed on Unipad?
Please ask the author if he has verified the effectiveness of the Unipad framework on other backbones.
For example resnet-50. If so, what can I do to verify that resnet-50 is pre-trained under unipad framework.