Du Tran

@kousik97 did you compile with CPU_ONLY := 1 option

For version 1.1, it is supported multi-GPUs. You need to install NCCL first (see instruction here https://github.com/NVIDIA/nccl), then un-comment this line (https://github.com/facebook/C3D/blob/master/C3D-v1.1/Makefile.config.example#L103), then compile.

Which prototxt file did you use for testing? Did you set, use_global_stats: true? If not, batch norm will not do anything with batch size =1 or not doing right if...

@ltnghia this may be due to the prediction size is too big (the same as input size, which is K x 16 x 112 x 112) which is 200704 K...

@rohitqut I don't understand your question. Can you explain a bit more on this? What are you trying to predict, if it is classification, then simply use 'prob' as your...

@deepaksharma36 thanks for helping. @sophiazy have you solved your issue? you don't need to scale your image in advanced, the data layer can do that for you (with a little...

It looks like you cannot read video from your list. It failed here (https://github.com/facebook/C3D/blob/master/C3D-v1.0/tools/compute_volume_mean_from_list.cpp#L92). Which mean size_in_datum = 0. Can you share your list file?

I don't really understand the issue here. If it run out of memory, you can just simple reduce the batch-size. Also 10,000 is a very large batch size.

@sainisanjay thanks for helping!

But that is a different architecture based on Resnet. For C3D-VGG, I have to convert it into the new format later. Will keep this issue open until I share with...