Model and proto files for python

[huyvvo]

Thank you for the python version of face parsing.
Could you also share the model and proto files for Python? I tried the model for matlab but it does not work in python with the following error message:

huy@Dom:~/Face_parsing_python/python_scripts$ python
Python 3.6.4 |Anaconda custom (64-bit)| (default, Jan 16 2018, 18:10:19)
[GCC 7.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.

import caffe
import test
ls
Traceback (most recent call last):
File "", line 1, in
NameError: name 'ls' is not defined
args = test.parse_arges()
Traceback (most recent call last):
File "", line 1, in
AttributeError: module 'test' has no attribute 'parse_arges'
args = test.parse_args()
args
Namespace(device_id=2, model='models/face_parsing_v1_iter_20800.caffemodel', proto='models/face_parsing_v1_test4one.prototxt', root_dir='/Data/')
net = caffe.Net(args.proto, args.model, caffe.TEST)
WARNING: Logging before InitGoogleLogging() is written to STDERR
W0623 13:34:52.418781 6606 _caffe.cpp:139] DEPRECATION WARNING - deprecated use of Python interface
W0623 13:34:52.418848 6606 _caffe.cpp:140] Use this instead (with the named "weights" parameter):
W0623 13:34:52.418864 6606 _caffe.cpp:142] Net('models/face_parsing_v1_test4one.prototxt', 1, weights='models/face_parsing_v1_iter_20800.caffemodel')
[libprotobuf ERROR google/protobuf/text_format.cc:298] Error parsing text-format caffe.NetParameter: 456:16: Message type "caffe.LayerParameter" has no field named "resize_param".
F0623 13:34:52.425428 6606 upgrade_proto.cpp:90] Check failed: ReadProtoFromTextFile(param_file, param) Failed to parse NetParameter file: models/face_parsing_v1_test4one.prototxt
*** Check failure stack trace: ***
Aborted (core dumped)

[aarif96]

You need to build the caffe from the project. She has defined her own layers in the project which will get built only if you build caffe from the project

[huyvvo]

By building caffe from the project, do you mean that I need to use the caffe that she has given instead of downloading from http://caffe.berkeleyvision.org/installation.html?

[lcybuzz]

I think you should use provided caffe_dev which contains new defined layers such as Resize Layer.
However, I still have some problems with proto and model files. Are they the same with the Matlab version? In test.py the output layer is named 'ae_prob', but it is not defined in the Matlab version proto. Besides, I use 'ae_label' as output and test some images, and the results seem not right.
Have you solved this issue?@vohuy93

[KeyuLi]

Can you train this model successfully?@lcybuzz

[lcybuzz]

No. I just transfer the code from Matlab to Python and make some tests. It seems this repository is not complete. @KeyuLi

[yanglinxiabuaaa]

Is there any other way to handle face parsing? @KeyuLi @lcybuzz

[KeyuLi]

I have the the same question, I'm looking for python code about face parsing, but I haven't found it. @yanglinxiabuaaa

[yanglinxiabuaaa]

Also looking for python code about face parsing . What do you do with face parsing? GAN? @KeyuLi

[KeyuLi]

I'm not studying in face parsing, I just want to use it for other directions. So do you find any other codes? @yanglinxiabuaaa

[yanglinxiabuaaa]

I'm in the same situation as you, there's no good code to solve right now. 发自网易邮箱手机版 在2018年09月06日 16:52，Keyu Li 写道： I'm not studying in face parsing, I just want to use it for other directions. So do you find any other codes? @yanglinxiabuaaa — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[lcybuzz]

It seems face parsing doesn't have many available codes. I just found one GuideSeg using two FCNs. But it only works on an old version Caffe. @KeyuLi @yanglinxiabuaaa

[yanglinxiabuaaa]

I also find it，Haven't tried yet！I am trying : Exemplar-Based-Face-Parsing. But Programming language is not Python，it use lua! @lcybuzz @KeyuLi

[lcybuzz]

Yes, this is a classic algorithm used by many papers as comparisons. But it is not based on DL. I wonder if the performance is good enough for using currently. How is your test results? @yanglinxiabuaaa

[KeyuLi]

OK, I will try this code GuideSeg later. @lcybuzz @yanglinxiabuaaa

[yanglinxiabuaaa]

There hasn't been a good result so far . Maybe I should try GuideSeg @lcybuzz