search

zju3dv / GIFT

Code for "GIFT: Learning Transformation-Invariant Dense Visual Descriptors via Group CNNs" NeurIPS 2019
GNU Affero General Public License v3.0
191 stars 17 forks source link

Finetuning on GL3D #3

Closed Smellly closed 3 years ago

Smellly commented 4 years ago

Could you share the yaml file about finetuning on the GL3D datasets ? I can't reproduce the pretrained model results on relative pose error in the paper. Thank you.

liuyuan-pal commented 4 years ago

Hi~Since the code for training on the dataset GL3D need to be further reorganized, we do not provide the training code for it. But recently, we are working on testing the GIFT on the IMW challenge (https://vision.uvic.ca/image-matching-challenge), so we plan to release the model used in this challenge.