support NotEqual layer without broadcasting

Author: minglu2019

include/caffe/layers/not_equal_layer.hpp

@@ -0,0 +1,38 @@
+#ifndef CAFFE_NOT_EQUAL_LAYER_HPP_
+#define CAFFE_NOT_EQUAL_LAYER_HPP_
+
+#include <vector>
+
+#include "caffe/blob.hpp"
+#include "caffe/layer.hpp"
+#include "caffe/proto/caffe.pb.h"
+
+namespace caffe {
+
+	// implement of Tensorflow Operator: https://www.tensorflow.org/api_docs/python/tf/math/not_equal
+
+
+template <typename Dtype> class NotEqualLayer : public Layer<Dtype> {
+public:
+  explicit NotEqualLayer(const LayerParameter &param)
+      : Layer<Dtype>(param) {}
+  virtual void Reshape(const vector<Blob<Dtype>*>& bottom,
+	  const vector<Blob<Dtype>*>& top);
+
+  virtual inline const char *type() const { return "NotEqual"; }
+  virtual inline int MinBottomBlobs() const { return 1; }
+  virtual inline int ExactNumTopBlobs() const { return 1; }
+
+protected:
+  virtual void Forward_cpu(const vector<Blob<Dtype> *> &bottom,
+                           const vector<Blob<Dtype> *> &top);
+  virtual void Backward_cpu(const vector<Blob<Dtype>*>& top,
+	  const vector<bool>& propagate_down, const vector<Blob<Dtype>*>& bottom) {
+	  NOT_IMPLEMENTED;
+  };
+};
+
+} // namespace caffe
+
+#endif // CAFFE_NOT_EQUAL_LAYER_HPP_
+

src/caffe/layers/not_equal_layer.cpp

@@ -0,0 +1,54 @@
+#include <math.h>
+#include <vector>
+
+#include "caffe/layers/not_equal_layer.hpp"
+
+namespace caffe {
+
+	template <typename Dtype>
+	void NotEqualLayer<Dtype>::Reshape(const vector<Blob<Dtype>*>& bottom,
+		const vector<Blob<Dtype>*>& top) {
+
+		Blob<Dtype>* comparand = (bottom.size() > 1) ? bottom[1] : this->blobs_[0].get();
+		vector<int> bottom_shape = bottom[0]->shape();
+
+		// case 1: bottom[1] is a scalar(bottom[0] may be a scalar)
+		if (comparand->num_axes() == 0) {}
+		// case 2: bottom[0] and bottom[1] are tensor and have the same dimension
+		else if (comparand->num_axes() == bottom[0]->num_axes()) {
+			for (int i = 0; i < bottom[0]->num_axes(); ++i) {
+				CHECK_EQ(bottom[0]->shape(i), comparand->shape(i)) << "Broadcasting is not supported now!!! Please confirm that 2 inputs have the same shape!!";
+			}
+		}
+		// case 3: bottom[0] and bottom[1] are tensor and have different dimension/shape
+		else {
+			CHECK_EQ(bottom[0]->num_axes(), comparand->num_axes()) << "Broadcasting is not supported now!!! Please confirm that 2 inputs have the same shape!!";
+		}
+		vector<int> top_shape = bottom_shape;
+		top[0]->Reshape(top_shape);
+	}
+
+	template <typename Dtype>
+	void NotEqualLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom,
+		const vector<Blob<Dtype>*>& top) {
+		const Dtype* bottom_data = bottom[0]->cpu_data();
+		Blob<Dtype>* comparand = (bottom.size() > 1) ? bottom[1] : this->blobs_[0].get();
+		const Dtype* comparand_data = comparand->cpu_data();
+		Dtype* top_data = top[0]->mutable_cpu_data();
+		if (comparand->num_axes() == 0) {
+			for (int i = 0; i < top[0]->count(); ++i) {
+				top_data[i] = bool(bottom_data[i] != comparand_data[0]);
+			}
+		}
+		else {
+			for (int i = 0; i < top[0]->count(); ++i) {
+				top_data[i] = bool(bottom_data[i] != comparand_data[i]);
+			}
+		}
+	}
+
+INSTANTIATE_CLASS(NotEqualLayer);
+REGISTER_LAYER_CLASS(NotEqual);
+
+} // namespace caffe
+