Fail to train customed data using train_deep_ls.py

[fangchuan]

Hi, thank you for release this fantastic work, I adapt your codes to train my data(warp from iGibson_obj). Everything is fine until I went into the train_deep_ls.py, It seems that python fail to handle DataLoader or anyother variable related to gradient when using multiprocess toolkit.

DeepLocalShapes/train_deep_ls.py

Lines 546 to 560 in 24ee928

	res = pool.map(functools.partial(trainer,
	sdf_tree = sdf_tree,
	sdf_grid_radius = sdf_grid_radius,
	lat_vecs = lat_vecs,
	sdf_data = sdf_data,
	indices = indices,
	cube_size = cube_size,
	outer_sum = outer_sum,
	outer_lock = outer_lock,
	decoder = decoder,
	loss_l1 = loss_l1,
	do_code_regularization = do_code_regularization,
	code_reg_lambda = code_reg_lambda,
	epoch = epoch),
	enumerate(sdf_grid_indices))

I havnt figure out how to solve this problem, could you help me? Please. @Kamysek @Freephi

There is the printed logging message:

My server used in this experiment is configured as below:
python: 3.6.13
torch: 1.4.0
cuda: 10.1
os: ubuntu 18.04

[sunyuanfu]

Hello.I wondr if you have solved your problem? I have trouble in setting up the environment too.