Open vvrmahendra opened 6 years ago
you should take a look at this :https://github.com/facebookresearch/Detectron/blob/master/tools/convert_pkl_to_pb.py
in here it says that FPN is not supported , so actually most models can't be converted , or is there a hack I don't know of ?
I think assert not cfg.FPN.FPN_ON, "FPN not supported."
means if cfg.FPN.PFN_ON is set to False, then it will print "FPN not supported". Actually FPN is supported except that the flag is set to False. @bendidi
Actually assert not cfg.FPN.FPN_ON, "FPN not supported."
means exactly the opposite of what you said :)
my first try was with FPN_on set on True
'FPN': {'COARSEST_STRIDE': 32,
'DIM': 256,
'EXTRA_CONV_LEVELS': False,
'FPN_ON': True,
'MULTILEVEL_ROIS': True,
'MULTILEVEL_RPN': True,
'ROI_CANONICAL_LEVEL': 4,
'ROI_CANONICAL_SCALE': 224,
'ROI_MAX_LEVEL': 5,
'ROI_MIN_LEVEL': 2,
'RPN_ANCHOR_START_SIZE': 32,
'RPN_ASPECT_RATIOS': (0.5, 1, 2),
'RPN_MAX_LEVEL': 6,
'RPN_MIN_LEVEL': 2,
'ZERO_INIT_LATERAL': False},
and I still get the assertion error :
Traceback (most recent call last):
File "tools/convert_pkl_to_pb.py", line 571, in <module>
main()
File "tools/convert_pkl_to_pb.py", line 525, in main
assert not cfg.FPN.FPN_ON, "FPN not supported."
AssertionError: FPN not supported.
Sorry, I misunderstood the code. You are right.@bendidi Hope someone else can explain it.
We are working on FPN support (the CollectAndDistributeFpnRpnProposals
operator) with standard Caffe2 C++, so stay tuned.
@orionr
Do you have a rough estimation of the timeline about this FPN support with standard Caffe2 c++? Thanks a ton!
I'm also interested in transfer the pkl file to pb. Want to know what is the current status on this. Thanks!
If #372 is merged it should close this issue.
I am facing issue in getting mask, lost mask in the conversion. I have used " Detectron/tools/convert_pkl_to_pb.py " file to convert my model.
How can i get mask as well?
Help appreciated, thanks :)
how to convert a .pb to .pkl format?
I tried to convert mask r-cnn pkl file to caffe2 pb file with the
convert_pkl_to_pb.py
script. I am confused what to pass in for --cfg option. There are several yaml files that are inconfig
directory, but it looks like these are for train, test etc. The documentation in theconvert_pkl_to_pb.py
says to copy / create a new yaml based oflib/core/config.py
but I am not sure what options would go in this yaml file. Is there a sample yaml config file available?