Models: Feature Request: Exporting .meta file of an already pre-trained model (.ckpt) in tf.slim
Using TF.slim and the pre-trained checkpoints I can evaluate the performance of a model. Also, using exporting inference graph I can freeze the model including its weights. However, I would like to find a way to export .meta file of a model as well. How can I do that using only the given the .ckpt file? Say on the inception_v3 case.
I guess something has to be changed in the export_inference_graph.py file to have this feature as an option.
amirjamez
All 3 comments
@sguada Can you comment on this feature request?
andydavis1
on 14 Dec 2017
@sguada, on the tf.slim models webpage, there are pre-trained models ready to use for training/evaluations. These files are the old .ckpt format, they lack the .meta and .index files (architecture of the model also referred to as .pbtxt). Although, using the provided python scripts on that page, one can directly perform training, evaluation, fine-tuning, and freezing, but the python script of each model is needed to manually construct the graph nodes if one is interested in looking at the weights as-is (without finetuning). Constructing the whole model on the newer version of TF requires 3 files: .meta, .index, and .data and I did not find any straight path to generate those (expect finetuning) to generate those three.
amirjamez
on 14 Dec 2017
Hi There,
We are checking to see if you still need help on this, as this seems to be considerably old issue. Please update this issue with the latest information, code snippet to reproduce your issue and error you are seeing.
If we don't hear from you in the next 7 days, this issue will be closed automatically. If you don't need help on this issue any more, please consider closing this.
tensorflowbutler
on 30 Jan 2020
Related issues
XavDCtpz
路
3Comments
chenyuZha
路
3Comments
jacknlliu
路
3Comments
atabakd
路
3Comments
twangnh
路
3Comments
Most helpful comment
@sguada, on the
tf.slimmodels webpage, there are pre-trained models ready to use for training/evaluations. These files are the old.ckptformat, they lack the.metaand.indexfiles (architecture of the model also referred to as.pbtxt). Although, using the provided python scripts on that page, one can directly perform training, evaluation, fine-tuning, and freezing, but the python script of each model is needed to manually construct the graph nodes if one is interested in looking at the weights as-is (without finetuning). Constructing the whole model on the newer version of TF requires 3 files:.meta,.index, and.dataand I did not find any straight path to generate those (expect finetuning) to generate those three.