search

gpcv-liujin / REDNet

implementation of REDNet (A Novel Recurrent Encoder-Decoder Structure for Large-Scale Multi-view Stereo Reconstruction from An Open Aerial Dataset)
53 stars 11 forks source link

Reg. Pretrained Model #5

Open deepak242424 opened 4 years ago

deepak242424 commented 4 years ago

Can you please share the links for pretrained models of your experiments in the paper?

gpcv-liujin commented 4 years ago

Can you please share the links for pretrained models of your experiments in the paper?

The pretrained model has been uploaded. You can get it from 'https://github.com/gpcv-liujin/REDNet/blob/master/MODEL_FOLDER/MODEL_FOLDER.zip'

deepak242424 commented 4 years ago

Thanks. I didn't notice that earlier.

I want to ask one more question, in your paper you showed the colmap results. Did you use ground truth extrinsic/instrinsic parameters to generate point clouds for WHU data?

And also if you can tell how are you visualizing the depth maps from COLMAP.

Many thanks.

gpcv-liujin commented 4 years ago

Thanks. I didn't notice that earlier.

I want to ask one more question, in your paper you showed the colmap results. Did you use ground truth extrinsic/instrinsic parameters to generate point clouds for WHU data?

And also if you can tell how are you visualizing the depth maps from COLMAP.

Many thanks.

Sorry I just noticed your reply.

I used ground truth camera parameters as input, so that the depth results obtained by colmap can be compared with the provided ground truth. The output depth maps are stored in the directory of "dense/stereo/depth_maps/" in the colmap project. The depth map results are stored in "bin" format, so I reloaded the "bin" file and visualized it using python.

deepak242424 commented 4 years ago

thanks for your response.

  1. I ran the pretrained model on test set. For a particular image the predicted depth map from REDNet and GT depth map looks similar. But the range of predicted Depth map is in (493.5945, 526.7905) and GT map is in (123,131). So do I need to do any post-processing to compute the L1 distance between these two maps?
  2. In Colmaps's "dense/stereo/depth_maps/" folder two depth maps get stored one is photometric and one is geometric. Can you please tell which one is you are using?