facebookresearch / DeepSDF

Learning Continuous Signed Distance Functions for Shape Representation
MIT License
1.4k stars 259 forks source link

Training data and training hardware #62

Closed nikwl closed 3 years ago

nikwl commented 3 years ago

Hello, and thank you for your great work!

Would you be able to give any indication as to the size of the training/testing set you used to generate the results shown in the paper? Did you use the same number of samples as are defined in the examples splits files (for chairs, lamps, plaes, sofas, and tables)? I ask because, for chairs especially, it looks like you only use about 20% of the shapenet data.

Also, which gpus were used for training? The paper says training was performed with 8 Nvidia gpus for 8 hours, but doesn't give any hint as to what specific Nvidia gpus were used.

Thank you!