alopezgit / desc Goto Github PK

Hello, I want to ask how to validate the performance during training. As different metrices are used for evaluation of training and test data, how we can include the validation data during training to see the error rate during training

When I run the following script ./pretrain_semantic_depth.sh VKITTI_ROOT_FOLDER KITTI_ROOT_FOLDER, the following error appears.

Traceback (most recent call last):
File "train.py", line 55, in
model.optimize_parameters()
File "/userhome/DESC/models/DESC_model.py", line 513, in optimize_parameters
self.backward_G()
File "/userhome/DESC/models/DESC_model.py", line 368, in backward_G
self.net_s2t.train()
AttributeError: 'DESCModel' object has no attribute 'net_s2t'

Is my path set incorrectly? In addition, do pretrain_depth.sh and pretrain_semantic_depth.sh share a checkpoints folder, or are they stored separately?@alopezgit

I'm so sorry to disturb you！

“Ground truth depth is linearly interpolated for visualization” , I don't know how to get such depth images.
Can you release your code for getting ground truth depth?

Looking forward to your reply, thank you very much！

Hi,
In your paper, I guess that your visualization for depth map uses cmap 'magma'.
I use the same cmap but get the following result.

It looks more bright and yellow. I'm not sure if I make mistake.
Therefore, could you tell how you visualize the depth map or provide the code?
Many thanks.