xxlong0 / cnmnet Goto Github PK

Hello,

According to your paper, you divided ScanNet scenes into 1000 for training and 600 for testing.

But I wonder why file train_plane_view3_scans0_999_interval2_error01.txt only has 882 scenes?

Hi, thanks for your code and paper.
I want to ask you one thing.
How many days does it take for training network？

Thanks for your nice work!
I cannot find any code about depth fusion in the repo.
Can you provide code to fuse depth maps to mesh?
Thanks a lot!

Thank you for releasing the code implementation.

I have a question about the results.

I use your pretrained weights and get pred_depth in 7scenes test set. I evaluate these depth maps using https://github.com/xxlong0/CNMNet/blob/master/utils/metric.py#L448, but I cannot obtain identical results as listed in the paper.

the results I got,
| <1.25 | <1.25^2 | <1.25^3 | abs.rel | sq.rel | rmse | remse log | scale. inv
| 0.481 | 0.799 | 0.934 | 0.283 | 0.214 | 0.563 | 0.338 | 0.326

I want to know more details:

1. How can I get the translation_gt in https://github.com/xxlong0/CNMNet/blob/master/utils/metric.py#L448
2. Do you scale the depth map to 640x480 before evaluation
3. It seems that you only evaluate part of images (https://github.com/xxlong0/CNMNet/blob/master/eval.py#L557)
   ...

Hi,
Could you elaborate on what is the algorithm you used to obtain the normal ground truth for these datasets? for example scannet/sun3d. Appreciate it a lot!

Hi! I'm confused of the parameters used when adding 'prob_loss'. Why is the weight 'alpha' set to 0.2 in the paper, but 1 in the code?
Since there is a function 'criterion_prob' in your code, did you use the prob_map_loss or other supervision information during training?

line 197 in train.py:

            prob_map_loss, prob_map_gt = criterion_prob(prob_map, idepth_refined, gt_disparity[:, 0, :, :, :])

            prob_loss = 5 * prob_loss_depth + prob_loss_minusmean  # + prob_map_loss

Thanks a lot!

I am not able to find the expected file.
I did once download the 7_scenes dataset from microsoft.
attaching the ss