search

lmb-freiburg / flownet2

FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networks
https://lmb.informatik.uni-freiburg.de/Publications/2017/IMKDB17/
Other
1k stars 318 forks source link

./scripts/run-flownet.py google/protobuf/text_format.cc:274] Error parsing text-format caffe.NetParameter: 40:22: Message type "caffe.LayerParameter" has no field named "augmentation_param". #73

Closed le-lshr1214 closed 7 years ago

le-lshr1214 commented 7 years ago

python ./scripts/run-flownet.py ./models/FlowNet2-c/FlowNet2-c_weights.caffemodel ./models/FlowNet2-c/FlowNet2-c_deploy.prototxt.template ./data/ChairsSDHom/data/test/t0/00000.png ./data/ChairsSDHom/data/test/t0/00001.png z.flo WARNING: Logging before InitGoogleLogging() is written to STDERR W1012 16:49:16.114919 11905 _caffe.cpp:139] DEPRECATION WARNING - deprecated use of Python interface W1012 16:49:16.114943 11905 _caffe.cpp:140] Use this instead (with the named "weights" parameter): W1012 16:49:16.114959 11905 _caffe.cpp:142] Net('/tmp/tmp7eQgwh', 1, weights='./models/FlowNet2-c/FlowNet2-c_weights.caffemodel') [libprotobuf ERROR google/protobuf/text_format.cc:274] Error parsing text-format caffe.NetParameter: 40:22: Message type "caffe.LayerParameter" has no field named "augmentation_param". F1012 16:49:16.116185 11905 upgrade_proto.cpp:88] Check failed: ReadProtoFromTextFile(param_file, param) Failed to parse NetParameter file: /tmp/tmp7eQgwh

nikolausmayer commented 7 years ago

Please check these related issues: https://github.com/lmb-freiburg/flownet2/issues?utf8=%E2%9C%93&q=augmentation_param

Your problem might come from not source-ing our shell script, or from conflicting Caffe installations on your machine.

le-lshr1214 commented 7 years ago

@nikolausmayer ,My caffe path is wrong. the error is solved,but still have other error. thanks!

Hanimiao commented 7 years ago

How do you solve this problem? I have the same problem. @le-lshr1214

rehanpunjwani commented 4 years ago

can't find my caffe_path. I have installed Caffe using ubuntu apt command! Please help me with this.

nikolausmayer commented 4 years ago

@Rehan-Rehman-Punjwani I have installed Caffe using ubuntu apt command The normal Caffe will not work. You need to compile the Caffe code in this repository (and that is then also your caffe_path).

rehanpunjwani commented 4 years ago

@nikolausmayer I am compiling it your Caffe on google colab and still getting the augmentation error!

nikolausmayer commented 4 years ago

@Rehan-Rehman-Punjwani The error Message type "caffe.LayerParameter" has no field named "augmentation_param" indicates that when you execute the network, it loads something that is not coming from our Caffe; probably something with Protobuf. This typically happens when there are multiple Caffe installations on a machine without a clean separation, but it may happen in other cases that I have not encountered myself.

rozos commented 4 years ago

Why your Caffe is different than the official Caffe? If every Caffe application comes with a tailored Caffe then ...

nikolausmayer commented 4 years ago

@rozos I understand and agree with the criticism. Multiple reasons.

Because it was 2016 and the spirit of making your own research code runnable by others with no effort was not yet strong in our group. Because we needed custom layers and Caffe being a C++ framework wasn't really as conducive to shipping your own layer code as, say, Python. Because while you are working on a paper, keeping up with upstream commits and API changes is the least of your concerns, as is post-release support (FlowNet2 was a research project, not a consumer product—we didn't get points for UX).

It's not really our fault that Caffe and its dependencies don't play nice with having multiple installations in the same environment, but we are aware of Caffe's problems, and they are one reason for my Docker image: https://github.com/lmb-freiburg/flownet2-docker

There are re-implementations of FlowNet2 in newer frameworks that do not have this issue, e.g. https://github.com/sampepose/flownet2-tf

rozos commented 4 years ago

Thanks for the reply. After all, Caffe is not an active project, so transition to TF was a good choice.

On 28/03/2020 13:39, Nikolaus Mayer wrote:

@rozos https://github.com/rozos I understand and agree with the criticism. Multiple reasons.

Because it was 2016 and the spirit of making your own research code runnable by others with no effort was not yet strong in our group. Because we needed custom layers and Caffe being a C++ framework wasn't really as conducive to shipping your own layer code as, say, Python. Because while you are working on a paper, keeping up with upstream commits and API changes is the least of your concerns, as is post-release support (FlowNet2 was a research project, not a consumer product—we didn't get points for UX).

It's not really our fault that Caffe and its dependencies don't play nice with having multiple installations in the same environment, but we are aware of Caffe's problems, and they are one reason for my Docker image: https://github.com/lmb-freiburg/flownet2-docker

There are re-implementations of FlowNet2 in newer frameworks that do not have this issue, e.g. https://github.com/sampepose/flownet2-tf

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/lmb-freiburg/flownet2/issues/73#issuecomment-605435039, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB7ZATUNFIYZ64RNMNYQIVLRJXOYHANCNFSM4D62Y3FA.

nikolausmayer commented 4 years ago

I like to think so, too :slightly_smiling_face: