Minor changes: delete TODO.txt; ran/fixed lint; better formatting, etc.

include/caffe/blob.hpp

@@ -138,13 +138,14 @@ class Blob {
  inline int channels() const { return LegacyShape(1); }
  /// @brief Deprecated legacy shape accessor length: use shape(2) instead.
  inline int length() const { return (num_axes() == 5) ? LegacyShape(2) : 1; }
- //inline int length() const { return LegacyShape(2); }
  /// @brief Deprecated legacy shape accessor height: use shape(3) instead.
- inline int height() const { return (num_axes() == 5) ? LegacyShape(3) : LegacyShape(2); }
- //inline int height() const { return LegacyShape(3); }
+ inline int height() const {
+ return (num_axes() == 5) ? LegacyShape(3) : LegacyShape(2);
+ }
  /// @brief Deprecated legacy shape accessor width: use shape(4) instead.
- inline int width() const { return (num_axes() == 5) ? LegacyShape(4) : LegacyShape(3); }
- //inline int width() const { return LegacyShape(4); }
+ inline int width() const {
+ return (num_axes() == 5) ? LegacyShape(4) : LegacyShape(3);
+ }
  inline int LegacyShape(int index) const {
  CHECK_LE(num_axes(), 5)
  << "Cannot use legacy accessors on Blobs with > 5 axes.";

include/caffe/layers/video_data_layer.hpp

@@ -15,8 +15,7 @@
 // an extension the std::pair which used to store image filename and
 // its label (int). now, a frame number associated with the video filename
 // is needed (second param) to fully represent a video segment
-struct triplet
-{
+struct triplet {
  std::string first;
  int second, third;
 };

include/caffe/util/im2col.hpp

@@ -7,7 +7,7 @@ template <typename Dtype>
 void im2col_nd_cpu(const Dtype* data_im, const int num_spatial_axes,
  const int* im_shape, const int* col_shape,
  const int* kernel_shape, const int* pad, const int* stride,
- const int* dilation, Dtype* data_col, const bool forced_3d=false);
+ const int* dilation, Dtype* data_col, const bool forced_3d = false);
 
 template <typename Dtype>
 void im2col_cpu(const Dtype* data_im, const int channels,
@@ -20,7 +20,7 @@ template <typename Dtype>
 void col2im_nd_cpu(const Dtype* data_col, const int num_spatial_axes,
  const int* im_shape, const int* col_shape,
  const int* kernel_shape, const int* pad, const int* stride,
- const int* dilation, Dtype* data_im, const bool forced_3d=false);
+ const int* dilation, Dtype* data_im, const bool forced_3d = false);
 
 template <typename Dtype>
 void col2im_cpu(const Dtype* data_col, const int channels,

include/caffe/util/io.hpp

@@ -5,6 +5,7 @@
 #include <iomanip>
 #include <iostream> // NOLINT(readability/streams)
 #include <string>
+#include <vector>
 
 #include "google/protobuf/message.h"
 

scripts/cpp_lint.py

@@ -1595,10 +1595,10 @@ def CheckCaffeAlternatives(filename, clean_lines, linenum, error):
 def CheckCaffeDataLayerSetUp(filename, clean_lines, linenum, error):
  """Except the base classes, Caffe DataLayer should define DataLayerSetUp
  instead of LayerSetUp.
- 
+
  The base DataLayers define common SetUp steps, the subclasses should
  not override them.
- 
+
  Args:
  filename: The name of the current file.
  clean_lines: A CleansedLines instance containing the file.
@@ -1610,6 +1610,7 @@ def CheckCaffeDataLayerSetUp(filename, clean_lines, linenum, error):
  if ix >= 0 and (
  line.find('void DataLayer<Dtype>::LayerSetUp') != -1 or
  line.find('void ImageDataLayer<Dtype>::LayerSetUp') != -1 or
+ line.find('void VideoDataLayer<Dtype>::LayerSetUp') != -1 or
  line.find('void MemoryDataLayer<Dtype>::LayerSetUp') != -1 or
  line.find('void WindowDataLayer<Dtype>::LayerSetUp') != -1):
  error(filename, linenum, 'caffe/data_layer_setup', 2,
@@ -1622,6 +1623,7 @@ def CheckCaffeDataLayerSetUp(filename, clean_lines, linenum, error):
  line.find('void Base') == -1 and
  line.find('void DataLayer<Dtype>::DataLayerSetUp') == -1 and
  line.find('void ImageDataLayer<Dtype>::DataLayerSetUp') == -1 and
+ line.find('void VideoDataLayer<Dtype>::DataLayerSetUp') == -1 and
  line.find('void MemoryDataLayer<Dtype>::DataLayerSetUp') == -1 and
  line.find('void WindowDataLayer<Dtype>::DataLayerSetUp') == -1):
  error(filename, linenum, 'caffe/data_layer_setup', 2,

src/caffe/blob.cpp

@@ -34,15 +34,8 @@ void Blob<Dtype>::Reshape(const int num, const int channels, const int height,
 
 template <typename Dtype>
 void Blob<Dtype>::Reshape(const vector<int>& shape) {
- //std::cout << "-----------------------------"<<std::endl;
- //for (int i=0; i<shape.size(); ++i)
- // std::cout<<"shape["<<i<<"]="<<shape[i]<<std::endl;
  vector<int> new_shape = shape;
  const bool force_3d_ = (shape.size() == 4) && false;
- //if (force_3d_)
- // std::cout<<"!!!!! force_3d_="<<force_3d_<<" !!!!!"<<std::endl;
- //else
- // std::cout<<"force_3d_="<<force_3d_<<std::endl;
  if (force_3d_)
  new_shape.insert(new_shape.begin() + 2, 1);
  CHECK_LE(new_shape.size(), kMaxBlobAxes);
@@ -64,9 +57,6 @@ void Blob<Dtype>::Reshape(const vector<int>& shape) {
  data_.reset(new SyncedMemory(capacity_ * sizeof(Dtype)));
  diff_.reset(new SyncedMemory(capacity_ * sizeof(Dtype)));
  }
- //std::cout << "----------"<<std::endl;
- //for (int i=0; i<shape_.size(); ++i)
- // std::cout<<"shape_["<<i<<"]="<<shape_[i]<<std::endl;
 }
 
 template <typename Dtype>
@@ -441,8 +431,7 @@ bool Blob<Dtype>::ShapeEquals(const BlobProto& other) {
  LegacyShape(-3) == other.length() &&
  LegacyShape(-2) == other.height() &&
  LegacyShape(-1) == other.width();
- }
- else if ((other.has_num() || other.has_channels() || other.has_height() ||
+ } else if ((other.has_num() || other.has_channels() || other.has_height() ||
  other.has_width()) && !other.has_length()) {
  // Using deprecated 4D Blob dimensions --
  // shape is (num, channels, height, width).
@@ -498,10 +487,11 @@ void Blob<Dtype>::CopyFrom(const Blob& source, bool copy_diff, bool reshape) {
 
 template <typename Dtype>
 void Blob<Dtype>::FromProto(const BlobProto& proto, bool reshape) {
+ bool is_legacy_C3D_proto = false;
  if (reshape) {
  vector<int> shape;
- if ((proto.has_num() || proto.has_channels() ||
- proto.has_height() || proto.has_width()) && proto.has_length()) {
+ if (proto.has_num() && proto.has_channels() && proto.has_height() &&
+ proto.has_width() && proto.has_length()) {
  // Using deprecated 5D Blob dimensions --
  // shape is (num, channels, length, height, width).
  shape.resize(5);
@@ -510,18 +500,31 @@ void Blob<Dtype>::FromProto(const BlobProto& proto, bool reshape) {
  shape[2] = proto.length();
  shape[3] = proto.height();
  shape[4] = proto.width();
- }
  // backward compatibility
- else if ((proto.has_num() || proto.has_channels() ||
- proto.has_height() || proto.has_width()) && !proto.has_length()) {
- // Using deprecated 4D Blob dimensions --
- // shape is (num, channels, height, width).
- shape.resize(5);
- shape[0] = proto.num();
- shape[1] = proto.channels();
- shape[2] = 1;
- shape[3] = proto.height();
- shape[4] = proto.width();
+ } else if (proto.has_num() && proto.has_channels() && proto.has_height() &&
+ proto.has_width() && !proto.has_length()) {
+ // a hacky way to support old proto generated by facebook/C3D
+ const int forgotten_dim_size = proto.diff_size() / (proto.num() *
+ proto.channels() * proto.height() * proto.width());
+ is_legacy_C3D_proto = (forgotten_dim_size > 1);
+ LOG(INFO) << "legacy C3D proto detected";
+ if (is_legacy_C3D_proto) {
+ shape.resize(5);
+ shape[0] = proto.num();
+ shape[1] = proto.channels();
+ shape[2] = proto.height();
+ shape[3] = proto.width();
+ shape[4] = forgotten_dim_size;
+ } else {
+ // Using deprecated 4D Blob dimensions --
+ // shape is (num, channels, height, width).
+ shape.resize(5);
+ shape[0] = proto.num();
+ shape[1] = proto.channels();
+ shape[2] = 1;
+ shape[3] = proto.height();
+ shape[4] = proto.width();
+ }
  } else {
  shape.resize(proto.shape().dim_size());
  for (int i = 0; i < proto.shape().dim_size(); ++i) {
@@ -540,9 +543,17 @@ void Blob<Dtype>::FromProto(const BlobProto& proto, bool reshape) {
  data_vec[i] = proto.double_data(i);
  }
  } else {
- CHECK_EQ(count_, proto.data_size());
- for (int i = 0; i < count_; ++i) {
- data_vec[i] = proto.data(i);
+ // normal case
+ if (!is_legacy_C3D_proto) {
+ CHECK_EQ(count_, proto.data_size());
+ for (int i = 0; i < count_; ++i) {
+ data_vec[i] = proto.data(i);
+ }
+ // binary proto file created by legacy C3D code
+ } else {
+ for (int i = 0; i < count_; ++i) {
+ data_vec[i] = proto.diff(i);
+ }
  }
  }
  if (proto.double_diff_size() > 0) {
@@ -551,7 +562,7 @@ void Blob<Dtype>::FromProto(const BlobProto& proto, bool reshape) {
  for (int i = 0; i < count_; ++i) {
  diff_vec[i] = proto.double_diff(i);
  }
- } else if (proto.diff_size() > 0) {
+ } else if (proto.diff_size() > 0 && !is_legacy_C3D_proto) {
  CHECK_EQ(count_, proto.diff_size());
  Dtype* diff_vec = mutable_cpu_diff();
  for (int i = 0; i < count_; ++i) {

src/caffe/layers/base_conv_layer.cpp

@@ -17,13 +17,11 @@ void BaseConvolutionLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
  channel_axis_ = bottom[0]->CanonicalAxisIndex(conv_param.axis());
  const int num_axes = bottom[0]->num_axes();
  int first_spatial_axis;
- if (num_axes == 5 && channel_axis_ == 1 &&
- bottom[0]->shape(2) == 1) {
+ if (num_axes == 5 && channel_axis_ == 1 && bottom[0]->shape(2) == 1) {
  forced_3d_ = true;
- first_spatial_axis = 3; // not 2
- num_spatial_axes_ = 2; // not 3
- }
- else {
+ first_spatial_axis = 3; // not 2
+ num_spatial_axes_ = 2; // not 3
+ } else {
  forced_3d_ = false;
  first_spatial_axis = channel_axis_ + 1;
  num_spatial_axes_ = num_axes - first_spatial_axis;
@@ -47,22 +45,21 @@ void BaseConvolutionLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
  CHECK(num_kernel_dims == 1 ||
  num_kernel_dims == num_spatial_axes_ ||
  num_kernel_dims == num_spatial_axes_ + 1)
- << "kernel_size must be specified once, or once per spatial dimension "
- << "(kernel_size specified " << num_kernel_dims << " times; "
- << num_spatial_axes_ << " spatial dims).";
+ << "kernel_size must be specified once, or once per spatial "
+ << "dimension (kernel_size specified " << num_kernel_dims
+ << " times; " << num_spatial_axes_ << " spatial dims).";
  } else {
  CHECK(num_kernel_dims == 1 || num_kernel_dims == num_spatial_axes_)
- << "kernel_size must be specified once, or once per spatial dimension "
- << "(kernel_size specified " << num_kernel_dims << " times; "
- << num_spatial_axes_ << " spatial dims).";
+ << "kernel_size must be specified once, or once per spatial "
+ << "dimension (kernel_size specified " << num_kernel_dims
+ << " times; " << num_spatial_axes_ << " spatial dims).";
  }
  if (num_kernel_dims == 1) {
  for (int i = 0; i < num_spatial_axes_; ++i) {
  kernel_shape_data[i] =
  conv_param.kernel_size(0);
  }
- }
- else if (num_kernel_dims == num_spatial_axes_) {
+ } else if (num_kernel_dims == num_spatial_axes_) {
  for (int i = 0; i < num_spatial_axes_; ++i) {
  kernel_shape_data[i] =
  conv_param.kernel_size(i);
@@ -71,7 +68,7 @@ void BaseConvolutionLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
  if (num_kernel_dims == num_spatial_axes_ + 1) {
  for (int i = 0; i < num_spatial_axes_; ++i) {
  kernel_shape_data[i] =
- conv_param.kernel_size(i + 1); // ignore the first kernel_size
+ conv_param.kernel_size(i + 1);  // ignore the first kernel_size
  }
  }
  }
@@ -113,7 +110,7 @@ void BaseConvolutionLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
  else if (num_stride_dims == num_spatial_axes_ )
  stride_data[i] = conv_param.stride(i);
  else if (num_stride_dims == num_spatial_axes_ + 1)
- stride_data[i] = conv_param.stride(i + 1); // ignore the first one
+ stride_data[i] = conv_param.stride(i + 1);  // ignore the first one
  CHECK_GT(stride_data[i], 0) << "Stride dimensions must be nonzero.";
  }
  }
@@ -152,7 +149,7 @@ void BaseConvolutionLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
  else if (num_pad_dims == num_spatial_axes_ )
  pad_data[i] = conv_param.pad(i);
  else if (num_pad_dims == num_spatial_axes_ + 1)
- pad_data[i] = conv_param.pad(i + 1); // ignore the first one
+ pad_data[i] = conv_param.pad(i + 1);  // ignore the first one
  }
  }
  // Setup dilation dimensions (dilation_).
@@ -182,7 +179,7 @@ void BaseConvolutionLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
  else if (num_dilation_dims == num_spatial_axes_ )
  dilation_data[i] = conv_param.dilation(i);
  else if (num_dilation_dims == num_spatial_axes_ + 1)
- dilation_data[i] = conv_param.dilation(i + 1); // ignore the first one
+ dilation_data[i] = conv_param.dilation(i + 1);  // ignore the first one
  }
  // Special case: im2col is the identity for 1x1 convolution with stride 1
  // and no padding, so flag for skipping the buffer and transformation.
@@ -305,7 +302,8 @@ void BaseConvolutionLayer<Dtype>::Reshape(const vector<Blob<Dtype>*>& bottom,
  if (reverse_dimensions()) {
  conv_input_shape_data[i] = top[0]->shape(channel_axis_ + i + forced_3d_);
  } else {
- conv_input_shape_data[i] = bottom[0]->shape(channel_axis_ + i + forced_3d_);
+ conv_input_shape_data[i] = bottom[0]->shape(channel_axis_ + i +
+ forced_3d_);
  }
  }
  // The im2col result buffer will only hold one image at a time to avoid

src/caffe/layers/cudnn_conv_layer.cpp

@@ -98,8 +98,6 @@ void CuDNNConvolutionLayer<Dtype>::Reshape(
  bottom_offset_ = this->bottom_dim_ / this->group_;
  top_offset_ = this->top_dim_ / this->group_;
  const bool forced_3d = this->forced_3d_;
- //std::cout << "cudnn_conv_layer.cpp: num_spatial_axes="<<this->num_spatial_axes_<<std::endl;
- //std::cout << "cudnn_conv_layer.cpp: forced_3d="<<forced_3d<<std::endl;
  const int height = bottom[0]->shape(this->channel_axis_ + 1 + forced_3d);
  const int width = bottom[0]->shape(this->channel_axis_ + 2 + forced_3d);
  const int height_out = top[0]->shape(this->channel_axis_ + 1 + forced_3d);
@@ -110,8 +108,6 @@ void CuDNNConvolutionLayer<Dtype>::Reshape(
  const int* stride_data = this->stride_.cpu_data();
  const int stride_h = stride_data[0];
  const int stride_w = stride_data[1];
- //std::cout << "cudnn_conv_layer.cpp: h="<<height<<", w="<<width<<", h_o="<<height_out<<
- // ", w_o="<<width_out<<", pad_h="<<pad_h <<std::endl;
 
  // Specify workspace limit for kernels directly until we have a
  // planning strategy and a rewrite of Caffe's GPU memory mangagement

src/caffe/layers/im2col_layer.cpp

@@ -18,10 +18,8 @@ void Im2colLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
  forced_3d_ = true;
  else
  forced_3d_ = false;
- //std::cout << "forced_3d_=" << forced_3d_ << std::endl;
  const int first_spatial_dim = channel_axis_ + 1 + forced_3d_;
  num_spatial_axes_ = input_num_dims - first_spatial_dim;
- //std::cout << "num_spatial_axes_=" << num_spatial_axes_ << std::endl;
  CHECK_GE(num_spatial_axes_, 1);
  vector<int> dim_blob_shape(1, num_spatial_axes_);
  // Setup filter kernel dimensions (kernel_shape_).
@@ -128,11 +126,6 @@ void Im2colLayer<Dtype>::Reshape(const vector<Blob<Dtype>*>& bottom,
  / stride_data[i] + 1;
  top_shape[channel_axis_ + i + 1 + forced_3d_] = output_dim;
  }
- //{
- //std::cout << "----------------------" << std::endl;
- //for (int i=0; i<top_shape.size(); ++i)
- // std::cout << "top_shape["<<i<<"]="<<top_shape[i]<<std::endl;
- //}
  top[0]->Reshape(top_shape);
  num_ = bottom[0]->count(0, channel_axis_);
  bottom_dim_ = bottom[0]->count(channel_axis_);

src/caffe/layers/pooling_layer.cpp

@@ -86,7 +86,6 @@ void PoolingLayer<Dtype>::Reshape(const vector<Blob<Dtype>*>& bottom,
  channels_ = bottom[0]->channels();
  height_ = bottom[0]->height();
  width_ = bottom[0]->width();
- //std::cout<<"c="<<channels_<<",h="<<height_<<",w="<<width_<<std::endl;
  if (global_pooling_) {
  kernel_h_ = bottom[0]->height();
  kernel_w_ = bottom[0]->width();