Model in training mode during demo.py

qinzheng93 / GeoTransformer

[CVPR2022] Geometric Transformer for Fast and Robust Point Cloud Registration

MIT License

677 stars 70 forks source link

Model in training mode during demo.py #116

Open braydenrudisill opened 2 months ago

braydenrudisill commented 2 months ago

Model.eval() needs to be added in the demo.py for the model be in evaluation mode. Being stuck in training mode enables this line of code which uses the ground truth transformation.

# 7 Random select ground truth node correspondences during training
if self.training:
    ref_node_corr_indices, src_node_corr_indices, node_corr_scores = self.coarse_target(
        gt_node_corr_indices, gt_node_corr_overlaps
    )

cduch9 commented 1 month ago

Thanks a lot! It would have saved me a lot of time if I had seen this post earlier.