search

PKU-ICST-MIPL / OPAM_TIP2018

Source code of our TIP 2018 paper "Object-Part Attention Model for Fine-grained Image Classification"
91 stars 32 forks source link

about generate heatmap file #10

Open GYxiaOH opened 5 years ago

GYxiaOH commented 5 years ago
  1. Can you share your deploy file? I see the output size of prepare_image(x) is w h c batchsize ,but the input size of caffe is batchsize chanel w h . so when i run demo ,dimension always wrong,I don't know why

  2. weights_LR = net.params('CAM_fc_car',1).get_data();% get the softmax layer of the network

    scores = net.forward({prepare_image(img)});% extract conv features online
activation_lastconv = net.blobs('CAM_conv').get_data();

what is CAM_fc_car' and CAM_conv

mumuhe commented 5 years ago

@GYxiaOH CAM_fc_car means the last fc layer of your model, and CAM_conv means the last conv layer of your model.