search

j96w / DenseFusion

"DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusion" code repository
https://sites.google.com/view/densefusion
MIT License
1.1k stars 300 forks source link

Segmentation results of LINEMOD dataset #152

Closed cvchanghao closed 4 years ago

cvchanghao commented 4 years ago

Hi, Thanks for sharing your work. I'm wondering how you got the segment results of LINMOD dataset. As we know that only one object's mask per frame. Even in 02 folder, the masks are still not enough. So how to train the SegNet with LINEMOD? Or you train one SegNet for each of the models?

j96w commented 4 years ago

Hi, yes, we train one SegNet for each model since the released LineMOD dataset only contains the pose label of one object in each video. We generate the segmentation ground truth by projecting the object model transformed by the ground truth pose to the image space. Since LineMOD is a small dataset, we also add synthetic data generated by sixd_toolkit. Basically, we randomly sampled the object pose and use sixd_toolkit to render the rgb image and use COCO as the background. Note that, these synthetic data are only used for the training of the segmentation model, the training of the pose estimation model didn't use these synthetic data.