What's the pretrained_settings for r2plus1d_18?

[XinyuLyu]

❓ Questions and Help

Please note that this issue tracker is not a help form and this issue will be closed.

We have a set of listed resources available on the website. Our primary means of support is our discussion forum:

• Discussion Forum
  Hi,
  I want to know the pretrained_settings for r2plus1d_18, which has not been implemented, including input_space,input_size,input_range,mean,std. In pytorch.pretrainedmodels, they implement resnet152, vgg19_bn, inceptionv4 with those pretrained_settings, which are necessary for TransformImage to transform the image. If include pretrained_settings, it will be more user-friendly for r3d_18,mc3_18,r2plus1d_18 models to extract vidoe frame features.Many thanks!
  

cc @bjuncek

[bjuncek]

Hi @XinyuLyu, the parameters were as follows:

pretraining dataset: Kinetics 400 (June 2018 snapshot)
input_space: RGB
input_size: 3x16x112x112
mean=[0.43216, 0.394666, 0.37645]
std=[0.22803, 0.22145, 0.216989]
num_classes: 400

I will add README.md to the video classification models to make this clear :)

Thanks for your question,
Bruno

[XinyuLyu]

Thank you so much!!!

[fmassa]

Also note that those models were trained on 64 GPUs. Further details are in #2723