search

JiayuYANG / Self-supervised-CVP-MVSNet

Self-supervised Learning of Depth Inference for Multi-view Stereo (CVPR 2021)
86 stars 12 forks source link

About the pretrained model #6

Open Ingram9 opened 2 years ago

Ingram9 commented 2 years ago

Hello! Nice work here! Is the model provided in this code trained by self-supervision? It seems that the model provided here has the same name as the one provided in CVP-MVSNet. Thanks!

XYZ-qiyh commented 2 years ago

The model in self_training.sh is unsupervised/model_000010.ckpt. But the model provided is pretrained/model_000027.ckpt. `

Ingram9 commented 2 years ago

The model in self_training.sh is unsupervised/model_000010.ckpt. But the model provided is pretrained/model_000027.ckpt. `

Thanks for your reply. So the unsupervised model is not released,or it is pretrained/model_000027.ckpt,but the name is unchanged?

alexrich021 commented 2 years ago

Hi there,

Great work! Just following up on this, as I have the same question. Is the model provided in checkpoints/pretrained/model_000027.ckpt the final model from the self-training with pseudo depth map generation? Or is it the result of training with ground truth depth information?

Thanks!