just modified model parameters, no core code touched

Author: zhirongw

examples/imagenet/train_caffenet.sh

@@ -1,5 +1,6 @@
 #!/usr/bin/env sh
 
+GOOGLE_LOG_DIR="models/bvlc_reference_caffenet" \
 ./build/tools/caffe train \
-    --solver=models/bvlc_reference_caffenet/solver.prototxt
-    --log_dir=.
+    --solver=models/bvlc_reference_caffenet/solver.prototxt \
+    --gpu=1

examples/imagenet/train_googlenet.sh

@@ -2,4 +2,5 @@
 
 GOOGLE_LOG_DIR=models/googlenet \
 ./build/tools/caffe train \
-    --solver=models/googlenet/solver.prototxt
+    --solver=models/googlenet/solver.prototxt \
+    --gpu=2

examples/imagenet/train_googlenet_gpus.sh

@@ -0,0 +1,6 @@
+#!/usr/bin/env sh
+
+GLOG_alsologtostderr=1 \
+GOOGLE_LOG_DIR=models/googlenet \
+./build/examples/parallel/gpus.bin \
+    models/googlenet/solver.prototxt 1:2

examples/imagenet/train_vgg.sh

@@ -0,0 +1,6 @@
+#!/usr/bin/env sh
+
+GOOGLE_LOG_DIR=models/vgg \
+./build/tools/caffe train \
+    --solver=models/vgg/solver.prototxt \
+    --gpu=3

models/bvlc_reference_caffenet/solver.prototxt

@@ -1,15 +1,14 @@
 net: "models/bvlc_reference_caffenet/train_val.prototxt"
 test_iter: 1000
 test_interval: 1000
-base_lr: 0.01
+base_lr: 0.004
 lr_policy: "step"
 gamma: 0.1
 stepsize: 100000
 display: 20
 max_iter: 450000
-momentum: 0.9
+momentum: 0.95
 weight_decay: 0.0005
 snapshot: 10000
 snapshot_prefix: "models/bvlc_reference_caffenet/caffenet_train"
 solver_mode: GPU
-device_id: 2

models/bvlc_reference_caffenet/train_val.prototxt

@@ -7,7 +7,7 @@ layers {
   data_param {
     source: "/home/common/imagenet/ilsvrc12_train_lmdb"
     backend: LMDB
-    batch_size: 256
+    batch_size: 128
   }
   transform_param {
     crop_size: 227

models/googlenet/solver.prototxt

@@ -1,16 +1,15 @@
 net: "models/googlenet/train_val.prototxt"
 test_iter: 1000
 test_interval: 1000
-base_lr: 0.01
+base_lr: 0.001
 lr_policy: "step"
 gamma: 0.1
 stepsize: 100000
 display: 20
 max_iter: 450000
-momentum: 0.9
+momentum: 0.975
 weight_decay: 0.0005
 snapshot: 10000
 snapshot_prefix: "models/googlenet/googlenet_train"
 solver_mode: GPU
-device_id: 3
 test_initialization: true

models/googlenet/train_val.prototxt

@@ -106,7 +106,7 @@ layers {
     }
     bias_filler {
       type: "constant"
-      value: 1
+      value: 0
     }
   }
 }
@@ -136,7 +136,7 @@ layers {
     }
     bias_filler {
       type: "constant"
-      value: 1
+      value: 0
     }
   }
 }