crs4/SliceNet

The prediction is bad

Closed this issue · 1 comments

I have tried to load and evaluate the images in Matterport following the README, but the prediction is not good. I wonder if you share the wrong resnet50_m3d.pth by mistake. I use the image from "Matterport3D/1LXtFkjw3qL/pano_color/0b22fa63d0f54a529c525afbf2e8bb25.jpg" and here is the prediction:
1_depth_pred
n.

Hi Leo, i've tried the image you mentioned, you just need to use the right mask. in this case is also more important since a larger part of the scene is an outdoor. next i'll upload a pytorch dataset loader to propely load image, mask and ground truth.
Here attached the prediction using the mask:

Figure_0
d