Test-Time Augmentation (TTA) on MSeg3D for nuScenes testing set

[giakhanh2711]

Thank you very much for your outstanding work. We are currently experimenting with your MSeg3D model. According to your paper, the Test-Time Augmentation (TTA) proposed in SDSeg3D is used to evaluate MSeg3D on the nuScenes test set. However, I noticed that MSeg3D utilizes both LiDAR and image data, whereas it appears that SDSeg3D does not use image data.

I am not sure if I have missed something, but I have been unable to find the implementation for the TTA of MSeg3D on the nuScenes test set in the code. Therefore, I am encountering difficulties in experimenting with MSeg3D on this test set. Although I attempted to implement it myself, the results did not match those reported in the paper.

Could you please provide some guidance on how to implement TTA for MSeg3D which uses both LiDAR point cloud and images data?

Thank you so much for your assistance.