Closed xinliu111 closed 5 years ago
Is it an error when restoring my pre-trained model?
python src/train_sibling.py config/finetune.py 加载预训练模型没有问题,训练完之后的模型导入不了
Are you trying to continue training your own model that has already been fine-tuned? If that is the case, try to comment out all the items in the "replace_scopes" in config/finetune.py.
I get the same issue whether or not I try continuing from my own model or your base one. I'm still running an older version which does not have a "replace_scopes" item in the config (the other difference is I'm still back on the version that wants exactly 2 images per folder named 1.jpg and 2.jpg). This worked several months ago for me, all I did was generate some new data to train on and now I get the "biases not found in checkpoint"
Doh!!!! Somebody who shares this machine with me had downgraded to Tensorflow 1.1. I went back to 1.5 and training from your base model works.....trying to start from my pretrained sibling network came up with a "weights not found in checkpoints" issue.
Doh!!!! Somebody who shares this machine with me had downgraded to Tensorflow 1.1. I went back to 1.5 and training from your base model works.....trying to start from my pretrained sibling network came up with a "weights not found in checkpoints" issue.
@huckw Did you solve the problem?
训练后的模型导入报错??? NotFoundError (see above for traceback): Key FaceResNet/conv1/layer_0/Conv/biases not found in checkpoint