After having no luck with various manual build manuals for Caffe2 and OpenCV (within my anaconda python 2.7 environment, Ubuntu 16.04), I followed the advice to install the nightly build binaries via:
conda install pytorch-nightly torchvision -c pytorch
I want to use models from R2Plus1D for extracting action recognition features. However, when running their extract_features.py it will throw the following error message:
Traceback (most recent call last):
File "lib/extract_features.py", line 333, in <module>
main()
File "lib/extract_features.py", line 328, in main
ExtractFeatures(args)
File "lib/extract_features.py", line 131, in ExtractFeatures
devices=gpus,
File "/anaconda2/lib/python2.7/site-packages/caffe2/python/data_parallel_model.py", line 34, in Parallelize_GPU
Parallelize(*args, **kwargs)
File "/anaconda2/lib/python2.7/site-packages/caffe2/python/data_parallel_model.py", line 219, in Parallelize
input_builder_fun(model_helper_obj)
File "lib/extract_features.py", line 106, in input_fn
use_local_file=args.use_local_file,
File "/models/c3d/R2Plus1D/lib/utils/model_helper.py", line 120, in AddVideoInput
data, label, video_id = model.net.VideoInput(
File "/anaconda2/lib/python2.7/site-packages/caffe2/python/core.py", line 2171, in __getattr__
",".join(workspace.C.nearby_opnames(op_type)) + ']'
AttributeError: Method VideoInput is not a registered operator. Did you mean: []
Some issues mention that there is a simple solution. Does somebody know a solution?