I had the same problem. Based on this thread , I was able to solve this using the old Proto syntax as suggested. For example, I had to do this:
Rename the definition of layers from layers to layer . All types of layers are renamed with caffe documentation (or proto file examples) - for example, a layer of type: CONVOLUTION - type: "Convolution" , etc. Replace the new syntax:
blobs_lr: 1 blobs_lr: 1 weight_decay: 1 weight_decay: 0
for
param { name: "conv1_w" lr_mult: 1 decay_mult: 1 } param { name: "conv1_b" lr_mult: 2 decay_mult: 0 }
Now parsing and working with new drawings are wonderful. See the sample .prototxt files in the caffe package for a better intuition of what the working proto syntax looks like.
source share