I test the pre-trained model on another dataset and output the different images. However, I find that the img pose loss and img expression loss are always zero. Are there some problems?
The loss of one epoch:
error 0.5489,
reconstruction loss 0.4762,
val pose img pose loss(target<->pose img) 0.0000, pose img exp loss(input <->pose img) 0.0000
val exp img pose loss(input <->exp img) 0.0000, exp img exp loss(target<->exp img) 0.0000
val out img pose loss(input <->exp img) 0.0000, out img exp loss(target<->exp img) 0.0000
val exp img recon loss(input <->exp img) 0.0276, pose img recon loss(input <->pose img) 0.0408
val exp flow tv loss 0.0042
epoch:-1, current val total_error: 0.54893630743
the test result:
from lest to right: source image, target image, output image, expression image, pose image
In addtion, the learned representations and reconstructed images seem strange sometimes, have you obtained a better result?
I test the pre-trained model on another dataset and output the different images. However, I find that the img pose loss and img expression loss are always zero. Are there some problems?
The loss of one epoch: error 0.5489, reconstruction loss 0.4762, val pose img pose loss(target<->pose img) 0.0000, pose img exp loss(input <->pose img) 0.0000 val exp img pose loss(input <->exp img) 0.0000, exp img exp loss(target<->exp img) 0.0000 val out img pose loss(input <->exp img) 0.0000, out img exp loss(target<->exp img) 0.0000 val exp img recon loss(input <->exp img) 0.0276, pose img recon loss(input <->pose img) 0.0408 val exp flow tv loss 0.0042 epoch:-1, current val total_error: 0.54893630743
the test result:
from lest to right: source image, target image, output image, expression image, pose image
In addtion, the learned representations and reconstructed images seem strange sometimes, have you obtained a better result?