The image translation network translates the color from the exemplar to a projection of the 3D shape and the part segmentation from the projection to the exemplar.
Please download pre-trained VGG models form here, move them to ./models.
To train this network, please run this script.
python ./src/image_translation/train.py --name=rendered --dataset_mode=rendered --dataroot=./data/training_data/image_translation --niter=25 --niter_decay=25 --use_attention --maskmix --noise_for_mask --mask_epoch=25 --warp_mask_losstype=direct --PONO --PONO_C --vgg_normal_correct --batchSize=2 --gpu_ids=0 --checkpoints_dir=[checkpoint dir] --nThreads=0We thank CocosNet for their great work!