selective search window proposals and better batching of images

Makefile

@@ -63,7 +63,7 @@ TEST_BINS := ${TEST_OBJS:.o=.testbin}
 # Derive include and lib directories
 ##############################
 CUDA_INCLUDE_DIR := $(CUDA_DIR)/include
-CUDA_LIB_DIR := $(CUDA_DIR)/lib64 $(CUDA_DIR)/lib
+CUDA_LIB_DIR := $(CUDA_DIR)/lib $(CUDA_DIR)/lib64
 MKL_INCLUDE_DIR := $(MKL_DIR)/include
 MKL_LIB_DIR := $(MKL_DIR)/lib $(MKL_DIR)/lib/intel64
 

data/get_mnist.sh

@@ -15,4 +15,9 @@ gunzip train-labels-idx1-ubyte.gz
 gunzip t10k-images-idx3-ubyte.gz
 gunzip t10k-labels-idx1-ubyte.gz
 
+echo "Creating leveldb..."
+
+../examples/convert_mnist_data.bin train-images-idx3-ubyte train-labels-idx1-ubyte mnist-train-leveldb
+../examples/convert_mnist_data.bin t10k-images-idx3-ubyte t10k-labels-idx1-ubyte mnist-test-leveldb
+
 echo "Done."

data/lenet.prototxt

@@ -3,7 +3,7 @@ layers {
   layer {
     name: "mnist"
     type: "data"
-    source: "data/mnist-train-leveldb"
+    source: "mnist-train-leveldb"
     batchsize: 64
     scale: 0.00390625
   }
@@ -114,7 +114,7 @@ layers {
 }
 layers {
   layer {
-    name: "prob"
+    name: "loss"
     type: "softmax_loss"
   }
   bottom: "ip2"

data/lenet_solver.prototxt

@@ -1,12 +1,27 @@
-train_net: "data/lenet.prototxt"
-test_net: "data/lenet_test.prototxt"
+# The training protocol buffer definition
+train_net: "lenet.prototxt"
+# The testing protocol buffer definition
+test_net: "lenet_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: 0.01
+momentum: 0.9
+weight_decay: 0.0005
+# The learning rate policy
 lr_policy: "inv"
 gamma: 0.0001
 power: 0.75
+# Display every 100 iterations
 display: 100
-max_iter: 5000
-momentum: 0.9
-weight_decay: 0.0005
-test_iter: 100
-test_interval: 500
+# The maximum number of iterations
+max_iter: 10000
+# snapshot intermediate results
+snapshot: 5000
+snapshot_prefix: "lenet"
+# solver mode: 0 for CPU and 1 for GPU
+solver_mode: 1

data/lenet_test.prototxt

@@ -3,7 +3,7 @@ layers {
   layer {
     name: "mnist"
     type: "data"
-    source: "data/mnist-test-leveldb"
+    source: "mnist-test-leveldb"
     batchsize: 100
     scale: 0.00390625
   }

data/train_mnist.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env sh
+
+GLOG_logtostderr=1 ../examples/train_net.bin lenet_solver.prototxt

examples/convert_imageset.cpp

@@ -46,7 +46,7 @@ int main(int argc, char** argv) {
     LOG(INFO) << "Shuffling data";
     std::random_shuffle(lines.begin(), lines.end());
   }
-  LOG(INFO) << "A total of " << lines.size() << "images.";
+  LOG(INFO) << "A total of " << lines.size() << " images.";
 
   leveldb::DB* db;
   leveldb::Options options;

examples/finetune_net.cpp

@@ -19,9 +19,6 @@ int main(int argc, char** argv) {
     return 0;
   }
 
-  Caffe::SetDevice(0);
-  Caffe::set_mode(Caffe::GPU);
-
   SolverParameter solver_param;
   ReadProtoFromTextFile(argv[1], &solver_param);
 

examples/train_net.cpp

@@ -20,9 +20,6 @@ int main(int argc, char** argv) {
     return 0;
   }
 
-  Caffe::SetDevice(0);
-  Caffe::set_mode(Caffe::GPU);
-
   SolverParameter solver_param;
   ReadProtoFromTextFile(argv[1], &solver_param);