ethnhe / FFB6D

[CVPR2021 Oral] FFB6D: A Full Flow Bidirectional Fusion Network for 6D Pose Estimation.
MIT License
295 stars 72 forks source link

Setting up dataset with fused & rendered data #7

Closed CandiceD17 closed 3 years ago

CandiceD17 commented 3 years ago

While I trained this model with Linemod, I encountered the following warning:

Warning: Trainnig without fused data will hurt model performance Please generate fused data from https://github.com/ethnhe/raster_triangle.

I go to this github repo and compiled the rendered & fused data, but how can I incorporate these into the dataset of FFB6D? Thank you for your help~

ethnhe commented 3 years ago

Simply run the training scripts again and it works if the Warning disappears.

Specifically, if you soft link the LineMOD dataset Linemod_preprocessed to raster_triangle/Linemod_preprocessed using ln -s path_to_Linemod_preprocessed ./Linemod_preprocessed following the README, the generate rendered & fused data will be in Linemod_preprocessed/renders and Linemod_preprocessed/fuse. And If you soft link the same Linemod_preprocessed folder to ffb6d/datasets/linemod/Linemod_preprocessed. The scripts linemod_dataset.py will find these generated data.