1) I have pool 5 features in .npy format of size 512 numbers. How to input that as weights for input to a fully connected layer?
Input dimension of FC layer is 8192 which comes after pooling by flatenning 512 1 4 * 4 but here I have only 512. Do I need to typecast those constants as a tensor of this matrix dimensions and then use as input?
2) The inputs are images with read_clips_and_labels but my training input would be weights of that video with label id?? I have frozen the upper layers.
1) I have pool 5 features in .npy format of size 512 numbers. How to input that as weights for input to a fully connected layer? Input dimension of FC layer is 8192 which comes after pooling by flatenning 512 1 4 * 4 but here I have only 512. Do I need to typecast those constants as a tensor of this matrix dimensions and then use as input?
2) The inputs are images with read_clips_and_labels but my training input would be weights of that video with label id?? I have frozen the upper layers.