diff --git a/c/segNet.cpp b/c/segNet.cpp
index 4943c2ed0..7670a29f5 100644
--- a/c/segNet.cpp
+++ b/c/segNet.cpp
@@ -735,6 +735,19 @@ bool segNet::Process( void* image, uint32_t width, uint32_t height, imageFormat
 	return true;
 }
 
+bool segNet::GetClassScores( float** class_scores, uint32_t* width, uint32_t* height, uint32_t* num_classes )
+{
+	if ( mOutputs[0].CPU == NULL )
+	{
+		return false;
+	}
+	*class_scores = (float*) mOutputs[0].CPU;
+	*width = DIMS_W(mOutputs[0].dims);
+	*height = DIMS_H(mOutputs[0].dims);
+	*num_classes = DIMS_C(mOutputs[0].dims);
+	return true;
+}
+		
 
 // argmax classification
 bool segNet::classify( const char* ignore_class )
diff --git a/c/segNet.h b/c/segNet.h
index 21f702202..126de5ff9 100644
--- a/c/segNet.h
+++ b/c/segNet.h
@@ -234,6 +234,16 @@ class segNet : public tensorNet
 	bool Process( float* input, uint32_t width, uint32_t height, const char* ignore_class="void" );
 
 	/**
+	 * Return per-pixel class probabilities, as well as number of classes and size of the output layer.
+	 * Does not perform a memory copy.
+	 * @param class_scores float pointer to destination array
+	 * @param width pointer to the variable that will hold the output layer width
+	 * @param height pointer to the variable that will hold the output layer height
+	 * @param num_classes pointer to the variable that will hold the number of classes
+	 */
+	 bool GetClassScores( float** class_scores, uint32_t* width, uint32_t* height, uint32_t* num_classes );
+  
+  /**
 	 * Produce a colorized segmentation mask.
 	 */
 	template<typename T> bool Mask( T* output, uint32_t width, uint32_t height, FilterMode filter=FILTER_LINEAR )				{ return Mask((void*)output, width, height, imageFormatFromType<T>(), filter); }