I3D Resnet-50 feature extraction

how to extract the features of the 5th layer of the 50-layer
I3D residual neural network (ResNet-50) [7]. The feature is denoted by F ∈ Rb×c×n/2×w×h, where b, c, w and h indicate the batch size, number of channels, width and height respectively.

Reference paper : GLNet: Global Local Network for Weakly Supervised Action Localization

Take a look at:

1 Like

Instead, if you start from a TFHUB model it is not officially supported but there was a workaround:

1 Like