Fix a warning and add some more error handling.

dlib/dnn/loss.h

@@ -852,6 +852,11 @@ namespace dlib
             const float* out_data  = output_tensor.host();
             for (long i = 0; i < output_tensor.num_samples(); ++i, ++truth)
             {
+                const long long num_label_categories = truth->size();
+                DLIB_CASSERT(output_tensor.k() == num_label_categories,
+                    "Number of label types should match the number of output channels. "
+                    "output_tensor.k(): " << output_tensor.k() 
+                    << ", num_label_categories: "<< num_label_categories);
                 for (long k = 0; k < output_tensor.k(); ++k)
                 {
                     const float y = (*truth)[k];

dlib/dnn/loss_abstract.h

@@ -767,8 +767,9 @@ namespace dlib
                 - sub.get_output().nc() == 1
                 - sub.get_output().num_samples() == input_tensor.num_samples()
                 - sub.sample_expansion_factor() == 1
-                - all values pointed to by truth are std::vectors of non-zero elements.
-                  Nominally they should be +1 or -1, each indicating the desired class label.
+                - truth points to training_label_type elements, each of size sub.get_output.k().
+                  The elements of each truth training_label_type instance are nominally +1 or -1,
+                  each representing a binary class label.
         !*/
 
     };

dlib/test/dnn.cpp

@@ -3391,9 +3391,11 @@ namespace
             {
                 for (size_t j = 0; j < labels[i].size(); ++j)
                 {
-                    if (labels[i][j] == 1 && preds[i][j] < 0 ||
-                        labels[i][j] == 0 && preds[i][j] > 0)
+                    if ((labels[i][j] == 1 && preds[i][j] < 0) ||
+                        (labels[i][j] == 0 && preds[i][j] > 0))
+                    {
                         ++num_wrong;
+                    }
                 }
             }
             return num_wrong / labels.size() / dims;