vsitzmann/light-field-networks

class-specific custom datasets

xyzalanix opened this issue · 2 comments

xyzalanix commented

👋 Any pointers on how class-specific datasets are structured, or how they were created... or to the urls for those? atm I can't find them in the shared drive folder. Thank you!

HaoRiziyou commented

I want to reproduce the experiment of class-specific.
"python experiment_scripts/train_nmr.py --data_root=path_to_nmr_dataset"
Anyone knows which file in the google shared drive is used to be the data_root, because I unzip the file in google drive
but it dies not work.
Error:
File "/content/light-field-networks/hdf5_dataio.py", line 176, in get_num_instances
return len(instances)
TypeError: object of type 'int' has no len()

vsitzmann commented

Indeed - I forgot to upload the class-specific datasets! I uploaded them now!

For the class-specific training, we use the "train_single_class.py" script - this uses the hdf5_dataio.py file, which loads the single-class hdf5 dataset files (cars_train.hdf5, for instance).

For the multi-class training, we use the "train_nmr.py" script, which uses the "multiclass_dataio.py" file. Here, the data_root should be the root directoy that you unpacked the NMR files to (plus the intrinsics.txt file from the google drive!)

  • ❤️2
  • 🚀1