I have a question about the yaml. What is yaml' here? In caffe, models are save in .caffemodel, non yaml. I downloaded thepspnet101_VOC2012.caffemodelfrom @hszhao's repo and put it indata/modelsthen I ranpython convert.py -c data/models/pspnet101_VOC2012.caffemodel` but it generated following error:

Traceback (most recent call last):
  File "convert.py", line 161, in <module>
    main()
  File "/mnt/hdd1/cao/.global_env/local/lib/python2.7/site-packages/click/core.py", line 722, in __call__
    return self.main(*args, **kwargs)
  File "/mnt/hdd1/cao/.global_env/local/lib/python2.7/site-packages/click/core.py", line 697, in main
    rv = self.invoke(ctx)
  File "/mnt/hdd1/cao/.global_env/local/lib/python2.7/site-packages/click/core.py", line 895, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "/mnt/hdd1/cao/.global_env/local/lib/python2.7/site-packages/click/core.py", line 535, in invoke
    return callback(*args, **kwargs)
  File "convert.py", line 117, in main
    CONFIG = Dict(yaml.load(open(config)))
  File "/mnt/hdd1/cao/.global_env/local/lib/python2.7/site-packages/yaml/__init__.py", line 69, in load
    loader = Loader(stream)
  File "/mnt/hdd1/cao/.global_env/local/lib/python2.7/site-packages/yaml/loader.py", line 34, in __init__
    Reader.__init__(self, stream)
  File "/mnt/hdd1/cao/.global_env/local/lib/python2.7/site-packages/yaml/reader.py", line 85, in __init__
    self.determine_encoding()
  File "/mnt/hdd1/cao/.global_env/local/lib/python2.7/site-packages/yaml/reader.py", line 135, in determine_encoding
    self.update(1)
  File "/mnt/hdd1/cao/.global_env/local/lib/python2.7/site-packages/yaml/reader.py", line 165, in update
    exc.encoding, exc.reason)
yaml.reader.ReaderError: 'utf8' codec can't decode byte #xa2: invalid start byte
  in "data/models/pspnet101_VOC2012.caffemodel", position 2

I wonder whether a pretrained PSPnet is able to handle variable size input. I am using the ade20k model, and I noticed that the input image to the model is by default 473x473. And if I changed it to 512x512, then the segmentation mask would be somewhat off by a few pixels.

I wonder whether I should still input the image as size 473x473, but later using bilinear upsampling to enlarge the mask to 512x512?

Hi, thanks for your nice work! I'm using your code but get some error in model transform, these are my error info:

conv5_3_pool6_conv/bn -> ppm.stages.0.conv.bn.weight : Copied conv5_3_pool6_conv/bn -> ppm.stages.0.conv.bn.running_var : Copied conv5_3_pool6_conv/bn -> ppm.stages.0.conv.bn.bias : Copied conv5_3_pool6_conv/bn -> ppm.stages.0.conv.bn.running_mean : Copied conv5_4 -> final.conv5_4.conv.weight : Copied conv5_4/bn -> final.conv5_4.bn.weight : Copied conv5_4/bn -> final.conv5_4.bn.running_var : Copied conv5_4/bn -> final.conv5_4.bn.bias : Copied conv5_4/bn -> final.conv5_4.bn.running_mean : Copied conv6 -> final.conv6.weight : Copied conv6 -> final.conv6.bias : Copied conv4_24 -> aux.conv4_aux.conv.weight : Copied conv4_24/bn -> aux.conv4_aux.bn.weight : Copied conv4_24/bn -> aux.conv4_aux.bn.running_var : Copied conv4_24/bn -> aux.conv4_aux.bn.bias : Copied conv4_24/bn -> aux.conv4_aux.bn.running_mean : Copied Traceback (most recent call last): File "convert.py", line 168, in main() File "/home/wzj/opt/miniconda2/envs/seg/lib/python2.7/site-packages/click/core.py", line 722, in call return self.main(*args, **kwargs) File "/home/wzj/opt/miniconda2/envs/seg/lib/python2.7/site-packages/click/core.py", line 697, in main rv = self.invoke(ctx) File "/home/wzj/opt/miniconda2/envs/seg/lib/python2.7/site-packages/click/core.py", line 895, in invoke return ctx.invoke(self.callback, **ctx.params) File "/home/wzj/opt/miniconda2/envs/seg/lib/python2.7/site-packages/click/core.py", line 535, in invoke return callback(*args, **kwargs) File "convert.py", line 157, in main values = values.view_as(own_state[param_name]) File "/home/wzj/opt/miniconda2/envs/seg/lib/python2.7/site-packages/torch/tensor.py", line 198, in view_as return self.view(tensor.size()) RuntimeError: invalid argument 2: size '[21 x 256 x 3 x 3]' is invalid for input with 5376 elements at /pytorch/torch/lib/TH/THStorage.c:41

I'm using python3.5 and pytorch0.3.0

Could you please help me fix this, or just share the model file you have transformed, thanks alot!

when I convert the voc101 model to pytorch version , I test on VOC2012 val.txt, but the mean IoU is 79.6% , much lower than the author given which is 85.41%， is there any other details I ignored