update mnist

.gitignore

@@ -1 +1,5 @@
-log
+log
+mnist_train_lmdb
+mnist_test_lmdb
+*.caffemodel
+*.solverstate

mnist/get_feature.m

@@ -0,0 +1,37 @@
+caffe.set_mode_gpu();
+caffe.reset_all();
+net = caffe.Net('lenet_train_test.prototxt','lenet_iter_25000.caffemodel','train');
+batch_size = 128;
+all_feature = zeros(batch_size*100,3);
+all_label = zeros(batch_size*100,1);
+
+for i=1:100
+    f = net.forward({});
+    data_blob = net.blobs('ip1').get_data();
+    label_blob = net.blobs('label').get_data();
+    if size(data_blob,1) == 1 && size(data_blob,2) == 1
+        data_blob = reshape(data_blob,[size(data_blob,3) size(data_blob,4)]);
+    end;
+    all_feature((i-1)*batch_size+1:i*batch_size,:) = data_blob';
+    all_label((i-1)*batch_size+1:i*batch_size) = label_blob;
+end;
+% weight = reshape(net.blobs('id_weight').get_data(), [2,10]);
+
+cc = colormap(jet);
+close(1);
+figure(1);
+hold on;
+for l=0:9
+    scatter3(all_feature(all_label==l,1), all_feature(all_label==l,2),all_feature(all_label==l,3), ones(sum(all_label==l),1)*5,'MarkerFaceColor',cc(l * 6 + 1,:),'MarkerEdgeColor',cc(l * 6 + 1,:));
+end;
+legend('0','1','2','3','4','5','6','7','8','9');
+upper_bound = max(abs(all_feature(:))) * 1.2;
+
+% for l=0:9
+%     plot(weight(1,l+1) * upper_bound /2 , weight(2,l+1) * upper_bound / 2,'x','MarkerSize',10);
+% end;
+hold off;
+axis([-upper_bound upper_bound -upper_bound upper_bound]);
+
+
+% distance = pdist2(all_feature, weight');

mnist/lenet_solver.prototxt

@@ -0,0 +1,29 @@
+# The train/test net protocol buffer definition
+net: "lenet_train_test.prototxt"
+# test_iter specifies how many forward passes the test should carry out.
+# In the case of MNIST, we have test batch size 100 and 100 test iterations,
+# covering the full 10,000 testing images.
+test_iter: 100
+# Carry out testing every 500 training iterations.
+test_interval: 500
+# The base learning rate, momentum and the weight decay of the network.
+base_lr: 0.1
+momentum: 0.9
+#weight_decay: 0
+weight_decay: 0.0005
+# The learning rate policy
+lr_policy: "multistep"
+gamma: 0.1
+power: 0.75
+# Display every 100 iterations
+display: 100
+# The maximum number of iterations
+stepvalue: 15000
+stepvalue: 20000
+stepvalue: 25000
+max_iter: 25000
+# snapshot intermediate results
+snapshot: 5000
+snapshot_prefix: "lenet"
+# solver mode: CPU or GPU
+solver_mode: GPU