Chuhang Zou

@jackaceuser For perspective images with type prediction, the pooling operation is needed for shrinking down the parameters. But I agree with you for general case in solving panoramic/perspective images the...

@github163sl Please refer to this discussion: https://github.com/zouchuhang/LayoutNetv2/issues/1#issuecomment-545273935

@kazu0622 If you train with images of non-cuboid shape but with cuboid shape ground truth, your trained network will have cuboid predictions for all cases. I'd suggest you remove images...

@kazu0622 If your trained data are all cuboid shape, then it's still workable to predict non-cuboid corners and edges. You cannot just replace "pano_line_solver.m" with "pano_line_solver_6.m", since "pano_line_solver.m" assumes input...

@kazu0622 "panofull_joint_box_pretrained.t7" is trained with images of both cuboid and non-cuboid shape, but all with cuboid shape ground truth. Thus for inference, given non-cuboid shape, it will predict cuboid results...

@kazu0622 As mentioned in https://github.com/zouchuhang/LayoutNet/issues/25#issuecomment-458033375, you need to exclude the 3D parameter regressor in the model since there're non-cuboid layouts now.

@kazu0622 The Manhattan Line estimation is a preprocessing step and is independent on the 3D reconstruction step. To better estimate room shape, please follow https://github.com/zouchuhang/LayoutNet/issues/25#issuecomment-458033375 "1. Determine whether to generate...

@teasherm The box parameter stored in "panoContext_box_train.t7" are normalized to be zero mean and standard deviation, causing those negative scale factors. I include the preprocessing script in "preprocessPano.m", you can...

@jackaceuser You can refer to the ground-truth masking in Section 3.3, last sentence in the third paragraph of the paper(http://openaccess.thecvf.com/content_cvpr_2018/CameraReady/0409.pdf). Since our ground-truth layout edge and corner map contains >95%...

@lijing1996 The optimization step is sampling-based and the solver is L-BFGS based, therefore it allows a slight quantitative fluctuation in prediction, as long as you use the correct full pre-trained...