search

autonomousvision / convolutional_occupancy_networks

[ECCV'20] Convolutional Occupancy Networks
https://pengsongyou.github.io/conv_onet
MIT License
830 stars 113 forks source link

scannet.yaml #38

Closed BingCS closed 2 years ago

BingCS commented 2 years ago

Hi Pengsong,

Could you please send me the evaluation setting on Scannet (scannet.yaml)?

Many thanks in advance!

Best, Bing

pengsongyou commented 2 years ago

Hi Bing,

Since we directly use the model pretrained from the synthetic room dataset, you can simply reuse most parts of the config files. For example, you can use the grid64 model, and simply change two lines in https://github.com/autonomousvision/convolutional_occupancy_networks/blob/master/configs/pointcloud/room_grid64.yaml#L4-L5 to

classes: ['scenes']
path: data/ScanNet

Maybe you also want to change the generation_dir to generation_dir: generation_scannet.

Best, Songyou

BingCS commented 2 years ago

Thanks for your reply. May I know the values of unit_size, query_vol_size and resolution_0 when you evaluate in the sliding-window way?

Best, Bing

pengsongyou commented 2 years ago

For the scannet experiment, we did not evaluate it in the sliding-window manner, but normalize it into the unit cube and evaluate in the entire room level. However, you can definitely try the sliding-window evaluation out by yourself and I guess the results should be better.

BingCS commented 2 years ago

Great thanks!