Open luoshanwei opened 6 years ago
Hi @luoshanwei , Thank you for your interest in the repo.
In my case, I restore pre-trained model (Resnet) from --pre_trained_model
directory, and last layer is not restored by default (refer to here). On the other hand, trained checkpoints are restored from --model_dir
directory, which contains last layer. Thus does not need to use --not-restore-last flag.
I hope this answers your question.
@rishizek hello, I meet the same question,as follows: Key Variable not found in checkpoint [[Node: save/RestoreV2 = RestoreV2[dtypes=[DT_INT32], _device="/job:localhost/replica:0/task:0/device:CPU:0"](_arg_save/Const_0_0, save/RestoreV2/tensor_names, save/RestoreV2/shape_and_slices)]] I removed the tf.train.init_checkpoint,this problem still remains, I use inceptionV3 as the baseline, from your answer, checkpoint needs to be updated as well.so how to update the checkpoint? hope your answer,thanks
Hi @DRACOyu ,
The code in the repo initially restores the model's weights from pre_trained
model mentioned above. Then, the checkpoint is restored from here if exist. Please refer to official document for estimator for detail.
Perhaps your model_dir
contains previous different architecture model. You may be able to run model if you remove your model_dir
.
I find more detailed explanation in DrSleep's repo It say "If restoring weights from the PASCAL models for your dataset with a different number of classes, you will also need to pass the --not-restore-last flag, which will prevent the last layers of size 21 from being restored" where is the param "--not-restore-last" in your code Looking forward to your reply