The reproduction effect of the Hayao dataset significantly differs from the effects of the provided weights. #30
Description
I trained the model using the default parameters.
python3 train.py --anime_image_dir dataset/Hayao
--real_image_dir dataset/train_photo
--model v2 \ # animeGAN version, can be v1 or v2
--batch 8
--amp \ # Turn on Automatic Mixed Precision training
--init_epochs 10
--exp_dir runs
--save_interval 1
--gan_loss lsgan \ # one of [lsgan, hinge, bce]
--init_lr 1e-4
--lr_g 2e-5
--lr_d 4e-5
--wadvd 300.0\ # Aversarial loss weight for D
--wadvg 300.0\ # Aversarial loss weight for G
--wcon 1.5\ # Content loss weight
--wgra 3.0\ # Gram loss weight
--wcol 30.0\ # Color loss weight
--use_sn\ # If set, use spectral normalization, default is False
torch==2.0.0+cu118
torchvision== 0.15.1+cu118
color-transfer-py==0.0.4
The effects obtained differ greatly from the effects of the weights you provided.
To achieve similar effects, do I need to change the parameters or is there an issue with my environment?
The left figure below shows my experimental results.
color_transfer_py 0.0.4
torch 1.13.0+cu116
torchaudio 0.13.0+cu116
torchvision 0.14.0+cu116
The left figure below shows my experimental results.