CycleGAN
CycleGAN copied to clipboard
Starting torch is really slow
Hello, I'm currently on Ubuntu 16.04 and use cuda 7.5 and cudnn 5.1
The Cezanne Example works, but starting torch takes an unusual amount of time. When I start it, it goes until GPU Mode
and then nothing happens for about 5 Minutes. After that the program seems to resume and does everything as expected.
Any Idea where this long waiting period could come from?
`
DATA_ROOT=./datasets/ae_photos name=style_cezanne_pretrained model=one_direction_test phase=test loadSize=256 fineSize=256 resize_or_crop="scale_width" th test.lua
------------------- Options -------------------
DATA_ROOT: ./datasets/ae_photos
align_data: 0
aspect_ratio: 1
batchSize: 1
cache_dir: ./cache
checkpoints_dir: ./checkpoints
continue_train: 1
cudnn: 1
display: 1
display_id: 200
fineSize: 256
flip: 0
gpu: 1
how_many: all
input_nc: 3
loadSize: 256
model: one_direction_test
nThreads: 1
name: style_cezanne_pretrained
norm: instance
output_nc: 3
phase: test
resize_or_crop: scale_width
results_dir: ./results/
serial_batches: 1
test: 1
which_direction: AtoB
which_epoch: latest
GPU Mode
`
Not sure if it is related to this repo. Could you run other GPU jobs with Torch?