search

liuyuan-pal / Gen6D

[ECCV2022] Gen6D: Generalizable Model-Free 6-DoF Object Pose Estimation from RGB Images
GNU General Public License v3.0
596 stars 74 forks source link

Replacement of the detection network #82

Open EvdoTheo opened 1 year ago

EvdoTheo commented 1 year ago

Hello, I'm interested in knowing why you used the VGG-11 network to extract the feature maps instead of others, e.g. Resnet. Is it feasible to make this modification now? Thank you in advance!

liuyuan-pal commented 1 year ago

Hi, the target is to match images. Generally, VGG-like structure is better than ResNet for matching tasks. For example, widely used descriptor network SuperPoint uses a VGG-like structure.

EvdoTheo commented 1 year ago

Thanks for the clarification! Did you try other networks for the matching task except for VGG?