[egs] rotation for image augmentation in CIFAR example (#1955)

egs/cifar/v1/local/nnet3/run_resnet_1d.sh

@@ -0,0 +1,142 @@
+#!/bin/bash
+
+# 1d is as 1c but adding rotation in image augmentation.
+
+# local/nnet3/compare.sh exp/resnet1c_cifar10 exp/resnet1d_cifar10
+# System                 resnet1c_cifar10   resnet1d_cifar10
+# final test accuracy:        0.9514           0.9537      
+# final train accuracy:            1           0.9966
+# final test objf:         -0.157244          -0.139607
+# final train objf:      -0.00751868         -0.0219607
+# num-parameters:           1322730            1322730
+
+# local/nnet3/compare.sh exp/resnet1c_cifar100 exp/resnet1d_cifar100
+# System                 resnet1c_cifar100 resnet1d_cifar100
+# final test accuracy:       0.7627        0.7687
+# final train accuracy:       0.96         0.9276
+# final test objf:         -0.862205    -0.812203
+# final train objf:        -0.174973    -0.265734
+# num-parameters:           1345860       1345860
+# steps/info/nnet3_dir_info.pl exp/resnet1c_cifar10{,0}
+# exp/resnet1d_cifar10: num-iters=133 nj=1..2 num-params=1.3M dim=96->10 combine=-0.04->-0.03 loglike:train/valid[87,132,final]=(-0.153,-0.044,-0.022/-0.25,-0.173,-0.140) accuracy:train/valid[87,132,final]=(0.946,0.9880,0.9966/0.921,0.946,0.954)
+# exp/resnet1d_cifar100: num-iters=133 nj=1..2 num-params=1.3M dim=96->100 combine=-0.33->-0.29 loglike:train/valid[87,132,final]=(-0.81,-0.37,-0.27/-1.15,-0.95,-0.81) accuracy:train/valid[87,132,final]=(0.760,0.897,0.928/0.68,0.737,0.769)
+
+# Set -e here so that we catch if any executable fails immediately
+set -euo pipefail
+
+
+
+# training options
+stage=0
+train_stage=-10
+dataset=cifar10
+srand=0
+reporting_email=
+affix=1d
+
+
+# End configuration section.
+echo "$0 $@"  # Print the command line for logging
+
+. ./cmd.sh
+. ./path.sh
+. ./utils/parse_options.sh
+
+if ! cuda-compiled; then
+  cat <<EOF && exit 1
+This script is intended to be used with GPUs but you have not compiled Kaldi with CUDA
+If you want to use GPUs (and have them), go to src/, and configure and make on a machine
+where "nvcc" is installed.
+EOF
+fi
+
+
+
+dir=exp/resnet${affix}_${dataset}
+
+egs=exp/${dataset}_egs2
+
+if [ ! -d $egs ]; then
+  echo "$0: expected directory $egs to exist.  Run the get_egs.sh commands in the"
+  echo "    run.sh before this script."
+  exit 1
+fi
+
+# check that the expected files are in the egs directory.
+
+for f in $egs/egs.1.ark $egs/train_diagnostic.egs $egs/valid_diagnostic.egs $egs/combine.egs \
+         $egs/info/feat_dim $egs/info/left_context $egs/info/right_context \
+         $egs/info/output_dim; do
+  if [ ! -e $f ]; then
+    echo "$0: expected file $f to exist."
+    exit 1;
+  fi
+done
+
+
+mkdir -p $dir/log
+
+
+if [ $stage -le 1 ]; then
+  mkdir -p $dir
+  echo "$0: creating neural net configs using the xconfig parser";
+
+  num_targets=$(cat $egs/info/output_dim)
+
+  # Note: we hardcode in the CNN config that we are dealing with 32x3x color
+  # images.
+
+
+  nf1=48
+  nf2=96
+  nf3=256
+  nb3=128
+
+  a="num-minibatches-history=40.0"
+  common="$a required-time-offsets=0 height-offsets=-1,0,1"
+  res_opts="$a bypass-source=batchnorm"
+
+  mkdir -p $dir/configs
+  cat <<EOF > $dir/configs/network.xconfig
+  input dim=96 name=input
+  conv-layer name=conv1 $a height-in=32 height-out=32 time-offsets=-1,0,1 required-time-offsets=0 height-offsets=-1,0,1 num-filters-out=$nf1
+  res-block name=res2 num-filters=$nf1 height=32 time-period=1 $res_opts
+  res-block name=res3 num-filters=$nf1 height=32 time-period=1 $res_opts
+  conv-layer name=conv4 height-in=32 height-out=16 height-subsample-out=2 time-offsets=-1,0,1 $common num-filters-out=$nf2
+  res-block name=res5 num-filters=$nf2 height=16 time-period=2 $res_opts
+  res-block name=res6 num-filters=$nf2 height=16 time-period=2 $res_opts
+  conv-layer name=conv7 height-in=16 height-out=8 height-subsample-out=2 time-offsets=-2,0,2 $common num-filters-out=$nf3
+  res-block name=res8 num-filters=$nf3 num-bottleneck-filters=$nb3 height=8 time-period=4 $res_opts
+  res-block name=res9 num-filters=$nf3 num-bottleneck-filters=$nb3 height=8 time-period=4 $res_opts
+  res-block name=res10 num-filters=$nf3 num-bottleneck-filters=$nb3 height=8 time-period=4 $res_opts
+  channel-average-layer name=channel-average input=Append(2,6,10,14,18,22,24,28) dim=$nf3
+  output-layer name=output learning-rate-factor=0.1 dim=$num_targets
+EOF
+  steps/nnet3/xconfig_to_configs.py --xconfig-file $dir/configs/network.xconfig --config-dir $dir/configs/
+fi
+
+
+if [ $stage -le 2 ]; then
+
+  steps/nnet3/train_raw_dnn.py --stage=$train_stage \
+    --cmd="$train_cmd" \
+    --image.augmentation-opts="--horizontal-flip-prob=0.5 --horizontal-shift=0.1 --vertical-shift=0.1 --rotation-degree=30 --rotation-prob=0.5 --num-channels=3" \
+    --trainer.srand=$srand \
+    --trainer.max-param-change=2.0 \
+    --trainer.num-epochs=100 \
+    --egs.frames-per-eg=1 \
+    --trainer.optimization.num-jobs-initial=1 \
+    --trainer.optimization.num-jobs-final=2 \
+    --trainer.optimization.initial-effective-lrate=0.003 \
+    --trainer.optimization.final-effective-lrate=0.0003 \
+    --trainer.optimization.minibatch-size=256,128,64 \
+    --trainer.optimization.proportional-shrink=50.0 \
+    --trainer.shuffle-buffer-size=2000 \
+    --egs.dir="$egs" \
+    --use-gpu=true \
+    --reporting.email="$reporting_email" \
+    --dir=$dir  || exit 1;
+fi
+
+
+exit 0;

egs/cifar/v1/local/nnet3/run_resnet_1e.sh

@@ -0,0 +1,142 @@
+#!/bin/bash
+
+# 1e is as 1d but with more filters and epochs.
+
+# local/nnet3/compare.sh exp/resnet1d_cifar10/ exp/resnet1e_cifar10/
+# System                resnet1d_cifar10 resnet1e_cifar10
+# final test accuracy:       0.9537      0.9583
+# final train accuracy:       0.9966      0.9994
+# final test objf:         -0.139607   -0.124945
+# final train objf:       -0.0219607 -0.00603407
+# num-parameters:           1322730     3465194
+
+# local/nnet3/compare.sh exp/resnet1d_cifar100 exp/resnet1e_cifar100
+# System                resnet1d_cifar100 resnet1e_cifar100
+# final test accuracy:       0.7687      0.7914
+# final train accuracy:       0.9276     0.9922
+# final test objf:         -0.812203   -0.786857
+# final train objf:        -0.265734   -0.0514912
+# num-parameters:           1345860     3511364
+# steps/info/nnet3_dir_info.pl exp/resnet1c_cifar10{,0}
+# exp/resnet1e_cifar10: num-iters=186 nj=1..2 num-params=3.5M dim=96->10 combine=-0.01->-0.01 loglike:train/valid[123,185,final]=(-0.109,-0.026,-0.0060/-0.21,-0.167,-0.125) accuracy:train/valid[123,185,final]=(0.963,0.9936,0.9994/0.930,0.949,0.958)
+# exp/resnet1e_cifar100/: num-iters=186 nj=1..2 num-params=3.5M dim=96->100 combine=-0.09->-0.07 loglike:train/valid[123,185,final]=(-0.53,-0.109,-0.051/-1.06,-0.93,-0.79) accuracy:train/valid[123,185,final]=(0.844,0.9730,0.9922/0.713,0.760,0.791)
+
+# Set -e here so that we catch if any executable fails immediately
+set -euo pipefail
+
+
+
+# training options
+stage=0
+train_stage=-10
+dataset=cifar10
+srand=0
+reporting_email=
+affix=1e
+
+
+# End configuration section.
+echo "$0 $@"  # Print the command line for logging
+
+. ./cmd.sh
+. ./path.sh
+. ./utils/parse_options.sh
+
+if ! cuda-compiled; then
+  cat <<EOF && exit 1
+This script is intended to be used with GPUs but you have not compiled Kaldi with CUDA
+If you want to use GPUs (and have them), go to src/, and configure and make on a machine
+where "nvcc" is installed.
+EOF
+fi
+
+
+
+dir=exp/resnet${affix}_${dataset}
+
+egs=exp/${dataset}_egs2
+
+if [ ! -d $egs ]; then
+  echo "$0: expected directory $egs to exist.  Run the get_egs.sh commands in the"
+  echo "    run.sh before this script."
+  exit 1
+fi
+
+# check that the expected files are in the egs directory.
+
+for f in $egs/egs.1.ark $egs/train_diagnostic.egs $egs/valid_diagnostic.egs $egs/combine.egs \
+         $egs/info/feat_dim $egs/info/left_context $egs/info/right_context \
+         $egs/info/output_dim; do
+  if [ ! -e $f ]; then
+    echo "$0: expected file $f to exist."
+    exit 1;
+  fi
+done
+
+
+mkdir -p $dir/log
+
+
+if [ $stage -le 1 ]; then
+  mkdir -p $dir
+  echo "$0: creating neural net configs using the xconfig parser";
+
+  num_targets=$(cat $egs/info/output_dim)
+
+  # Note: we hardcode in the CNN config that we are dealing with 32x3x color
+  # images.
+
+
+  nf1=48
+  nf2=96
+  nf3=512
+  nb3=256
+
+  a="num-minibatches-history=40.0"
+  common="$a required-time-offsets=0 height-offsets=-1,0,1"
+  res_opts="$a bypass-source=batchnorm"
+
+  mkdir -p $dir/configs
+  cat <<EOF > $dir/configs/network.xconfig
+  input dim=96 name=input
+  conv-layer name=conv1 $a height-in=32 height-out=32 time-offsets=-1,0,1 required-time-offsets=0 height-offsets=-1,0,1 num-filters-out=$nf1
+  res-block name=res2 num-filters=$nf1 height=32 time-period=1 $res_opts
+  res-block name=res3 num-filters=$nf1 height=32 time-period=1 $res_opts
+  conv-layer name=conv4 height-in=32 height-out=16 height-subsample-out=2 time-offsets=-1,0,1 $common num-filters-out=$nf2
+  res-block name=res5 num-filters=$nf2 height=16 time-period=2 $res_opts
+  res-block name=res6 num-filters=$nf2 height=16 time-period=2 $res_opts
+  conv-layer name=conv7 height-in=16 height-out=8 height-subsample-out=2 time-offsets=-2,0,2 $common num-filters-out=$nf3
+  res-block name=res8 num-filters=$nf3 num-bottleneck-filters=$nb3 height=8 time-period=4 $res_opts
+  res-block name=res9 num-filters=$nf3 num-bottleneck-filters=$nb3 height=8 time-period=4 $res_opts
+  res-block name=res10 num-filters=$nf3 num-bottleneck-filters=$nb3 height=8 time-period=4 $res_opts
+  channel-average-layer name=channel-average input=Append(2,6,10,14,18,22,24,28) dim=$nf3
+  output-layer name=output learning-rate-factor=0.1 dim=$num_targets
+EOF
+  steps/nnet3/xconfig_to_configs.py --xconfig-file $dir/configs/network.xconfig --config-dir $dir/configs/
+fi
+
+
+if [ $stage -le 2 ]; then
+
+  steps/nnet3/train_raw_dnn.py --stage=$train_stage \
+    --cmd="$train_cmd" \
+    --image.augmentation-opts="--horizontal-flip-prob=0.5 --horizontal-shift=0.1 --vertical-shift=0.1 --rotation-degree=30 --rotation-prob=0.5 --num-channels=3" \
+    --trainer.srand=$srand \
+    --trainer.max-param-change=2.0 \
+    --trainer.num-epochs=140 \
+    --egs.frames-per-eg=1 \
+    --trainer.optimization.num-jobs-initial=1 \
+    --trainer.optimization.num-jobs-final=2 \
+    --trainer.optimization.initial-effective-lrate=0.003 \
+    --trainer.optimization.final-effective-lrate=0.0003 \
+    --trainer.optimization.minibatch-size=256,128,64 \
+    --trainer.optimization.proportional-shrink=50.0 \
+    --trainer.shuffle-buffer-size=2000 \
+    --egs.dir="$egs" \
+    --use-gpu=true \
+    --reporting.email="$reporting_email" \
+    --dir=$dir  || exit 1;
+fi
+
+
+exit 0;

src/nnet3bin/nnet3-egs-augment-image.cc

@@ -35,13 +35,17 @@ struct ImageAugmentationConfig {
   BaseFloat horizontal_flip_prob;
   BaseFloat horizontal_shift;
   BaseFloat vertical_shift;
+  BaseFloat rotation_degree;
+  BaseFloat rotation_prob;
   std::string fill_mode_string;
 
   ImageAugmentationConfig():
       num_channels(1),
       horizontal_flip_prob(0.0),
       horizontal_shift(0.0),
       vertical_shift(0.0),
+      rotation_degree(0.0),
+      rotation_prob(0.0),
       fill_mode_string("nearest") { }
 
 
@@ -57,6 +61,10 @@ struct ImageAugmentationConfig {
     po->Register("vertical-shift", &vertical_shift,
                  "Maximum allowed vertical shift as proportion of image "
                  "height.  Padding is with closest pixel.");
+    po->Register("rotation-degree", &rotation_degree,
+                 "Maximum allowed degree to rotate the image");
+    po->Register("rotation-prob", &rotation_prob,
+                 "Probability of doing rotation");
     po->Register("fill-mode", &fill_mode_string, "Mode for dealing with "
 		 "points outside the image boundary when applying transformation. "
 		 "Choices = {nearest, reflect}");
@@ -68,6 +76,8 @@ struct ImageAugmentationConfig {
                  horizontal_flip_prob <= 1);
     KALDI_ASSERT(horizontal_shift >= 0 && horizontal_shift <= 1);
     KALDI_ASSERT(vertical_shift >= 0 && vertical_shift <= 1);
+    KALDI_ASSERT(rotation_degree >=0 && rotation_degree <= 180);
+    KALDI_ASSERT(rotation_prob >=0 && rotation_prob <= 1);
     KALDI_ASSERT(fill_mode_string == "nearest" || fill_mode_string == "reflect");
   }
 
@@ -231,6 +241,14 @@ void PerturbImage(const ImageAugmentationConfig &config,
   // [ cos(theta)  -sin(theta)  0
   //   sin(theta)  cos(theta)   0
   //   0           0            1 ]
+  if (RandUniform() <= config.rotation_prob) {
+    BaseFloat theta = (2 * config.rotation_degree * RandUniform() -
+		       config.rotation_degree) / 180.0 * M_PI;
+    rotation_mat(0, 0) = cos(theta);
+    rotation_mat(0, 1) = -sin(theta);
+    rotation_mat(1, 0) = sin(theta);
+    rotation_mat(1, 1) = cos(theta);
+  }
 
   Matrix<BaseFloat> shear_mat(3, 3, kUndefined);
   shear_mat.SetUnit();