search

royorel / StyleSDF

Other
536 stars 50 forks source link

a problem when running train_volume_renderer.py #23

Open Ni1r opened 2 years ago

Ni1r commented 2 years ago

hi!thanks for your great work! when i am trying to train the volume render for my own datasets,i met an error "PIL.UnidentifiedImageError: cannot identify image file <_io.BytesIO object at 0x7f2f33fa74a0>".Before this i had followed the steps in "preparing your datasets" and transformed my dataset to lmdb format.By the way,i can get the sdf_init_models,then the error occurs.Can you give me some advice? Thanks!

royorel commented 2 years ago

Hi @Ni1r,

From the type of your error, it seems to me that the data loader is trying to load a non-image file. Try to print the file name when the error occurs to verify that.

Ni1r commented 2 years ago

Hi @Ni1r,

From the type of your error, it seems to me that the data loader is trying to load a non-image file. Try to print the file name when the error occurs to verify that.

Thanks, i will try later.Besides, can i see the structure of the datasets folder?

royorel commented 2 years ago

The original dataset (before running prepare data) should only contain images.

royorel commented 2 years ago

Your dataset can be at any resolution. The dataloader will resize the images to the volume renderer resolution. You can set this resolution with the flag --renderer_spatial_output_dim (default is 64). Keep in mind that if you change the default resolution you have to use this flag both in the volume renderer training and the full pipeline training.