search

edenton / svg

186 stars 56 forks source link

Reproducibility #4

Open awaelchli opened 5 years ago

awaelchli commented 5 years ago

Hi

I am not able to reproduce the result for KTH from your paper. The settings I used:

python train_svg_fp.py --batch_size 16 --dataset kth --image_width 64 --model vgg --g_dim 128 --z_dim 24 --beta 0.000001 --n_past 10 --n_future 10 --channels 1 --lr 0.0008 --data_root /path/to/data/ --log_dir /logs/will/be/saved/here/ The only change I made was the batch_size, which was set to 100 by default and did not fit into my memory.

Is there anything I have missed to properly reproduce your results?

ry85 commented 5 years ago

use train_svg_lp.py script (in fact its better to use LP model for all the three datasets)