search

yeephycho / nasnet-tensorflow

A nasnet in tensorflow
Apache License 2.0
136 stars 52 forks source link

fine tune all layers nasnet #2

Closed PythonImageDeveloper closed 6 years ago

PythonImageDeveloper commented 6 years ago

i want to do fine-tune all layers , but i dont know what do i do ? num my class is 5 when i dont use this flags flowing :
--checkpoint_exclude_scopes=final_layer,aux_11 \ --trainable_scopes=final_layer,aux_11

InvalidArgumentError (see above for traceback): Assign requires shapes of both tensors to match. lhs shape= [5] rhs shape= [1001] [[Node: save/Assign_1 = Assign[T=DT_FLOAT, _class=["loc:@aux_11/aux_logits/FC/biases"], use_locking=true, validate_shape=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](aux_11/aux_logits/FC/biases, save/RestoreV2_1)]]

yeephycho commented 6 years ago

Hi @zeynali , If you want to train all the layers, simply remove the "--trainable_scope=final_layer,aux_11", but you may find that your GPU memory is not large enough and receive a OOM warning.

You need to make modification to the class number before you start to finetune, follow the instruction and read carefully.

Yeephycho

PythonImageDeveloper commented 6 years ago

Is enough titan x for finetune nasnet-large all layers? Batch 32 . how many The author used GPUs for train large model?