summaryrefslogtreecommitdiff
path: root/models
AgeCommit message (Collapse)Author
2017-10-19fix learning ratejunyanz
2017-10-19update learning rate policyjunyanz
2017-10-19fix small issue in learning rate policyjunyanz
2017-10-10fix identity option conflict error, and also not specifying init_type in ↵Taesung Park
define_D
2017-10-07Update cycle_gan_model.pyJun-Yan Zhu
2017-10-06fix minor bugsjunyanz
2017-10-06support nc=1, add new leaerning rate policy and new initializationjunyanz
2017-09-23Update networks.pyJun-Yan Zhu
2017-09-22change the order of concanated channelsLambdaWill
To be consistent with Cyclegan(Torch), the latter half channels should receive the data from the skip connections.
2017-09-04Update networks.pyHungryof
2017-08-26fix the unaligned datasetjunyanz
2017-08-22initialized bias of conv to zero,hazirbas
deactivated bias for the conv layers which are followed by batch normalization.
2017-07-17changed the default option to use dropoutTaesung Park
2017-07-05add reflection padding layerjunyanz
2017-07-03fix instancenorm & batchnormjunyanz
2017-06-13predict single imagejunyanz
2017-06-12update dataset modejunyanz
2017-05-02use InstanceNorm2d from the latest pytorchjunyanz
2017-04-27changed the name of the loaded network from net_G_A to net_GTaesung Park
2017-04-271. Added one_direction_test_model that generates the outputs in only one ↵Taesung Park
direction 2. Changed the option naming from ntrain to max_dataset_size
2017-04-23fix which_direction bugjunyanz
2017-04-22add comment for python 2/3junyanz
2017-04-22add dropout option for Gjunyanz
2017-04-22fix the padding/kernel_size in Djunyanz
2017-04-21Following griegler's suggestion, fixed a bug in NLayerDiscriminator that ↵Taesung Park
always applied sigmoid to the output
2017-04-20add gpu_ids to save/cuda functionjunyanz
2017-04-18Fixed bugs in the Unet structureTaesung Park
2017-04-18fix assert bugjunyanz
2017-04-18add unet and update READMEjunyanz
2017-04-18fix assert bugjunyanz
2017-04-18first commitjunyanz