search

megvii-research / TransMVSNet

(CVPR 2022) TransMVSNet: Global Context-aware Multi-view Stereo Network with Transformers.
MIT License
262 stars 25 forks source link

How is this work compared with TransformerFusion? #6

Closed HaFred closed 2 years ago

HaFred commented 2 years ago

Didn't see any comparison in the paper. Hopefully, you could give us some insights, thank you very much.

DingYikang commented 2 years ago

The Transformerfusion is an implicit reconstruction method, which is mainly evaluated on indoor scene datasets (e.g., scannet, 7scenes). On the contrary, TransMVSNet reconstructs 3D models by estimating and fusing depth maps, which can be used to reconstruct arbitrary objects. Considering these differences between TransMVSNet and Transformerfusion, we didn't perform comparing.