AttributeError: Method VideoInput is not a registered operator. Did you mean: []

[953585130]

Hi, Thank you for the open source code provided! I have installed gpu support as per the tutorial, and I get an error when I execute the script that extracts features.

AttributeError: Method VideoInput is not a registered operator. Did you mean: []

I tried to debug or reported the same error, I hope to get your help!

[seelikat]

I get the same error after successfully installing caffe2 and pytorch in a local anaconda installation with:

conda install pytorch-nightly torchvision -c pytorch

Would be great to get help with this issue. I tried extracting C3D features with extract_features.py.

[953585130]

@kateiyas Thank you for your reply, but I didn't understand what you mean. Do you mean that you used anaconda to install caffe2 and pytorch and the same error occurred? But what should be done? I searched the Internet for a long time, probably said that it may be related to opencv, I tried to recompile the version but still have the same problem. I really don't know how to solve this problem.

[seelikat]

@953585130: Yes, I got the same error AttributeError: Method VideoInput is not a registered operator. Did you mean: []. The error was thrown after following the tutorial for feature extraction, calling extract_features.py (which I mention in my reply). I have also described it here.

[953585130]

@kateiyas I have solved this problem and I have re-downloaded other versions of pytorch, which supports VideoInput. You can try another version, the version given by the tutorial has a problem or it was not fully cloned when cloning.

[greathope]

@953585130 can you please tell me which version of pytorch you are using?

[seelikat]

@953585130 Thanks for leaving an update here that the problem can be solved. Can you provide us with more information on how you solved this issue, please?

By tutorial, did you mean you followed this installation tutorial?

It may also be useful for others stumbling over this thread to run the pytorch environment collection script provided by their devs and paste the output here.

[orionr]

The default binary builds are generated without OpenCV and ffmpeg dependency, which means that these operators (all of those under caffe2/video/) aren't built. You'll need to build from source to have these operators available.

We should update the R2Plus1 tutorial to make a note of this.

[953585130]

@greathope @kateiyas Starting with a new system, I installed cuda8.0 and cudnn6.0 directly after installing the graphics driver. The next steps I followed were tutorials installed step by step.

Follow the tutorial to clone pytorch, the entire pytorch is about 300MB. I searched online for a tutorial on installing caffe2, one of which has a pytorch zip file, which is 400MB after decompression. Then I compiled it with this and did not report an error when performing feature extraction.

I don't know if your email supports file transfer? If your mailbox supports file transfer function, please send me your full email address, I can send this compressed package to you.

Just orionr replied to this question, you can look at his explanation.

[greathope]

@953585130 could you please send that file to my email? my email is zhangmiaoatzju@gmail.com or zhangmiaoatzju@163.com, many thanks!!!

[953585130]

@greathope I have sent it to your 163 mailbox, you can try to solve your problem.

[greathope]

@953585130 I solved!!! thanks bro

[953585130]

@greathope Do you have qq, we can discuss it together.

[seelikat]

Just a note for people stumbling over this thread: As of to date, building pytorch from source with USE_FFMPEG=1 and USE_OPENCV=1 does not build the Caffe2 video operators that are necessary for R2Plus1D to work. Theoretically building pytorch (+caffe2) from source via USE_FFMPEG=1 USE_OPENCV=1 USE_LEVELDB=0 USE_LMDB=1 python setup.py install should solve the issue in the thread, but it doesn't; the same error still occurs. @dutran Could make sense to reopen this issue.

@953585130 Do you still have a link to the online tutorial you mention? Otherwise: muellerkatj gmail com – I would be very grateful for help in this matter.

[953585130]

@kateiyas The compressed file has been delivered to your mailbox, and the rest will be installed according to the tutorial.

[orionr]

Also, @kateiyas, a heads up that USE_FFMPEG wasn't passed through setup.py until https://github.com/pytorch/pytorch/pull/12543 landed, which was yesterday. If you could retest with latest master (git pull; git submodule update) that would be great, but feel free to use the compressed file as well.

[dutran]

@kateiyas can you try the latest pytorch master as mentioned by @orionr , if it does not work, we can re-open it. thanks.

[seelikat]

@dutran @orionr Thanks, the new master got me around the video operator bug.

(Running feature extraction still did not work because R2Plus1D will only work with GPU support, i.e. my NO_GPU=1 is not possible. If I compile with GPU support I get a segmentation fault now after compiling, coming from Caffe2 again. This is a separate issue already documented here.)

[shanshuo]

Hi @953585130 could you also send me the compressed pytorch file? shanshuo1992@gmail.com. I think it'll help a lot if you can upload it to the cloud, too. I'm struggling with the installation for several days. Thanks

[953585130]

@shanshuo Has been sent to your mailbox.

[lrd8715923]

@953585130 Hi could you also send me the compressed pytorch file.I can't build it sucess. my emil:424079971@qq.com Thanks.

[ChisamXz]

@953585130 Hi, could you also send me the pytorch file? I've really suffered enough from the installation. I'll really appreciate the sharing!

My email is: chisam.xiz@gmail.com

Thanks

[953585130]

@ChisamXz I will send it to your email later.

[yimikai]

@kateiyas I have solved this problem and I have re-downloaded other versions of pytorch, which supports VideoInput. You can try another version, the version given by the tutorial has a problem or it was not fully cloned when cloning.

@953585130 hi, bro, could you please send that file to my email? my email is 1544788527@qq.com or hudongfang06@163.com, many thanks!!!

[vinitmuchhala]

Hey @953585130 , can you send me an email as well? it is vinit.muchhala@hotmail.com thanks

[oLIVIa-Ld]

@953585130 hi, could you please send me an email for the correct pytorch? thank you, so much. oliviald@126.com

[Jinyong-Huang]

@953585130 I tried to install pytorch/caffe2 for a few days, but failed.could you please send that file to my email? my email is 543401923@qq.com

[R2D2oid]

I followed the tutorial instructions and after resolving a whole lot of other issues along the way still, get the infamous VideoInput error!

@953585130 @greathope @kateiyas or others, do you mind sharing a link to the built package? maybe on google drive or elsewhere?

[Weed-eat-meat]

@953585130 Hi，I had the same problrm, can you send me an email as well? it is syy960315@163.com. thanks

[syuqings]

@953585130 Hi, I had the same problem, can you send me an email for the correct pytorch as well? It is syuqing@ruc.edu.cn, thank you very much!

[NemioLipi]

@953585130 Hi, I am facing the same problem, would you please send me the correct pytorch version as well? My email address is nemiolipi@gmail.com,

[syuqings]

@NemioLipi Hi, I have solved the problem by getting back to pytorch version v1.1.0. The script is as follows:

git clone https://github.com/pytorch/pytorch.git cd pytorch && git checkout v1.1.0 && git submodule update --init --recursive

But to be noted, the 487 line of caffe2/video/video_input_op.h should be replaced with the following: ReinitializeTensor(&prefetchedlabel, vector(1, batchsize clip_pervideo multi_cropcount), at::dtype().device(CPU)); Otherwise it will throw errors when compiling.

Then, USE_OPENCV=1 USE_FFMPEG=1 USE_LMDB=1 USE_CUDA=1 python setup.py install

[nassimaNoufail]

@953585130 Hi, I have the same issue can you please email me the correct PyTorch version, my email: nassima.noufail@gmail.com Thank you!!

[nassimaNoufail]

I am getting the Video Input error, can you please send me the correct PyTorch version (my email : nassima.noufail@gmail.com or nnoufail@uh.edu ) @dutran @953585130 @kateiyas @greathope @shanshuo @ChisamXz @yimikai

[Tracy-Lee1993]

@953585130 Hi, I have the same issue can you please email me the correct PyTorch version，my email：lizhaoyang_1993@qq.com