example-inference-1 not running. Pre trained Model error

[PrinceP]

While running
python3 examples/example-inference-1.py --path-to-npz models/pretrained-models/models/openpose-plus/hao28-pose600000.npz --images data/mscoco2017/val2017/ --limit 1

Following error is encountered
File "/usr/local/lib/python3.5/dist-packages/numpy/lib/npyio.py", line 440, in load
return pickle.load(fid, **pickle_kwargs)
_pickle.UnpicklingError: invalid load key, 'v'.

[mandeman]

Is it because you are trying to pass the folder in --images instead of individual comma separated filenames?

[PrinceP]

@mandeman
I modified the script to run for the images in a folder.
I trained a new vggtiny model again for some iterations, used the script command
python3 examples/example-inference-1.py --path-to-npz models/pose30000.npz --base-model=vggtiny --images data/mscoco2017/val2017/ --limit 1
And got the output

3.933178 1 3933.177710 main
1.996399 1 1996.399403 create TfPoseEstimator
1.891422 1 1891.421556 e.inference
0.007528 1 7.527828 read_imgfile

But for uploaded hao28-pose600000.npz, its giving error.

[lgarithm]

what is the sha1sum of hao28-pose600000.npz on your computer?

mine is

0962749be446b10fd2843efcaab1cc5f3fd92273  hao28-pose600000.npz

[PrinceP]

b8bd3b50edb97d9c6abfbda7a0a56f75a3b7c334 hao28-pose600000.npz

Can you give the weights.

[lgarithm]

which means cat hao28-pose600000.npz yields

version https://git-lfs.github.com/spec/v1
oid sha256:3fdfd66b709653b62a7ac2906f4a4d94898e4bf909e376b91cde20e99041f905
size 34584656

Because that file is managed by git lfs

You can download the content of that file using the URL obtained from web:

https://github.com/tensorlayer/pretrained-models/raw/master/models/openpose-plus/hao28-pose600000.npz

[PrinceP]

Worked fine.
Had some questions about the inference benchmarks. Will open another issue.
Thanks

[karthee320]

I have downloaded the model referenced above and run the command

python3 examples/example-inference-1.py --path-to-npz=models/hao28-pose600000.npz --images=data/mscoco2017/val2017/000000000785.jpg --plot=True

But it is giving errors.

tensorflow.python.framework.errors_impl.InvalidArgumentError: Dimension 0 in both shapes must be equal, but are 7 and 3. Shapes are [7,7,128,128] and [3,3,128,128]. for 'Assign' (op: 'Assign') with input shapes: [7,7,128,128], [3,3,128,128].

I tried all the three model name(vgg, vggtiny,mobilenet) as Command Line argument for base-model. One of the dimensions were not equal. Can anyone know what is the reason?

[akhalili]

same problem above

[oddfox19]

@karthee320 and @akhalili
Found that I was able to get example-inference-1 working when i used the model name 'hao28_experimental'