- Do you have other download sources?
- We are currently tidying up the data for public release, if you need an early access for academic purposes, please send an email to [email protected].
- How to evaluate on scanobjectnn with model trained on ModelNet40?
- Please see the file "evaluate_real_trained_on_synthetic.py". You can see the class mapping file in "mapping2.py".
- What is the difference between main_split,split1,split2,split3 and split4?
- "Main_split" was used for the experiments in our main paper, while the other splits (1-4) are additional splits that we reported in our supplementary materials.
- H5 labels: Can you confirm me that the label-integer correspondence (in h5 splits) is the one from 'shape_names_ext.txt' in ascending order and starting from zero?
e.g.
{bag: 0, bin: 1, box: 2, cabinet: 3, chair: 4, desk: 5, display: 6, door: 7, shelf: 8, table: 9, bed: 10, pillow: 11, sink: 12, sofa: 13, toilet: 14}
- Yes, that is correct.
- I want to assure that you use -1 to denote background in the mask.
- Yes, that's right.
training_data
Folders and files
| Name | Name | Last commit date | ||
|---|---|---|---|---|
parent directory.. | ||||