Chuhang Zou

@JonathanLehner You can refer to the discussion below: https://github.com/zouchuhang/LayoutNetv2/issues/1#issuecomment-545273935

@jackureu There's a corrected mapping from the .t7 file to the panoContext in the /data folder. The original mapping for the Stanford 2d-3d file shall be consistent. For the rotation...

@afad-dataset That's defined in the LSUN toolkit, as we include the toolkit in line 2. You need to ask the LSUN indoor layout contest holder to get the toolkit.

@NikoCiminelliTufts Thank you for your interest in our work. I believe the ".mat" file you mentioned is the 3D box room parameter. We actually have an upgraded LayoutNet v2 architecture:...

@qinxiaoyu222 We train & test our network on a single NVIDIA Titan X GPU. You shall be able to run it on a GPU with 12GB memory.

@qinxiaoyu222 I think you encounter the "out of memory" issue, our model use nearly all the 12GB memory for better performance. You can try down sampling the resolution of the...

@zoubochao I checked "driver_pano_joint.lua" and it needed around 10.969GB to run, nearly touch to your 11GB memory limitation. If you don't want to down sample the resolution, it's better to...

> You can try down sampling the resolution of the input image by a factor of 2 and change the related network structure to fit this resolution. In this way...

@GalAvineri I check the total number of parameters of my model, this has the same number of parameters as yours. The memory cost differs from different deepnet tool. You can...

@luben2018 The printed "gradnorm" shows zero gradients. I guess your training get stuck, rerun the code is the simplest strategy to solve.