Fix for LSTM Diplopia issue

src/lstm/recodebeam.cpp

@@ -40,6 +40,11 @@ const int RecodeBeamSearch::kBeamWidths[RecodedCharID::kMaxCodeLen + 1] = {
 
 static const char *kNodeContNames[] = {"Anything", "OnlyDup", "NoDup"};
 
+// the minimum diplopia key is the minimum score (key) from
+// the network output to qualify as a likely 'real' character
+// for the purposes of identifying possible diplopia
+static const float kMinDiplopiaKey = 0.25;
+
 // Prints debug details of the node.
 void RecodeNode::Print(int null_char, const UNICHARSET &unicharset, int depth) const {
   if (code == null_char) {
@@ -65,6 +70,9 @@ RecodeBeamSearch::RecodeBeamSearch(const UnicharCompress &recoder, int null_char
     , beam_size_(0)
     , top_code_(-1)
     , second_code_(-1)
+    , in_possible_diplopia_(false)
+    , first_diplopia_code_(-1)
+    , second_diplopia_code_(-1)
     , dict_(dict)
     , space_delimited_(true)
     , is_simple_text_(simple_text)
@@ -182,7 +190,7 @@ RecodeBeamSearch::combineSegmentedTimesteps(
 
 void RecodeBeamSearch::calculateCharBoundaries(std::vector<int> *starts, std::vector<int> *ends,
                                                std::vector<int> *char_bounds_, int maxWidth) {
-  char_bounds_->push_back(0);
+  char_bounds_->push_back((*starts)[0]); 
   for (int i = 0; i < ends->size(); ++i) {
     int middle = ((*starts)[i + 1] - (*ends)[i]) / 2;
     char_bounds_->push_back((*ends)[i] + middle);
@@ -567,8 +575,8 @@ void RecodeBeamSearch::ExtractPathAsUnicharIds(const std::vector<const RecodeNod
       }
       rating -= cert;
     }
-    starts.push_back(t);
     if (t < width) {
+      starts.push_back(t);
       int unichar_id = best_nodes[t]->unichar_id;
       if (unichar_id == UNICHAR_SPACE && !certs->empty() && best_nodes[t]->permuter != NO_PERM) {
         // All the rating and certainty go on the previous character except
@@ -582,16 +590,18 @@ void RecodeBeamSearch::ExtractPathAsUnicharIds(const std::vector<const RecodeNod
       }
       unichar_ids->push_back(unichar_id);
       xcoords->push_back(t);
-      do {
-        double cert = best_nodes[t++]->certainty;
+      t++;
+      while (t < width && best_nodes[t]->duplicate) {
+        double cert = best_nodes[t]->certainty;
         // Special-case NO-PERM space to forget the certainty of the previous
         // nulls. See long comment in ContinueContext.
         if (cert < certainty ||
             (unichar_id == UNICHAR_SPACE && best_nodes[t - 1]->permuter == NO_PERM)) {
           certainty = cert;
         }
         rating -= cert;
-      } while (t < width && best_nodes[t]->duplicate);
+        t++;
+      }
       ends.push_back(t);
       certs->push_back(certainty);
       ratings->push_back(rating);
@@ -657,20 +667,46 @@ void RecodeBeamSearch::ComputeTopN(const float *outputs, int num_outputs, int to
       }
     }
   }
+  float top_key = 0.0F;
+  float second_key = 0.0F;
+  bool found_first_code = false;
+  bool found_second_code = false;
   while (!top_heap_.empty()) {
     TopPair entry;
     top_heap_.Pop(&entry);
+    if (in_possible_diplopia_ && entry.data() == first_diplopia_code_)
+      found_first_code = true;
+    if (in_possible_diplopia_ && entry.data() == second_diplopia_code_)
+      found_second_code = true;
     if (top_heap_.size() > 1) {
       top_n_flags_[entry.data()] = TN_TOPN;
     } else {
       top_n_flags_[entry.data()] = TN_TOP2;
       if (top_heap_.empty()) {
         top_code_ = entry.data();
+        top_key = entry.key();
       } else {
         second_code_ = entry.data();
+        second_key = entry.key();
       }
     }
   }
+  // need to identify if we are in a potential diplopia situation
+  // or if we already are, then determine if it is ended
+  if (in_possible_diplopia_) {
+    if (!found_first_code && !found_second_code){
+      in_possible_diplopia_ = false;
+      first_diplopia_code_ = -1;
+      second_diplopia_code_ = -1;
+    }
+  }
+  if (!in_possible_diplopia_) {
+    if (top_code_ != null_char_ && second_code_ != null_char_ && top_key > kMinDiplopiaKey && second_key > kMinDiplopiaKey){
+      in_possible_diplopia_ = true;
+      first_diplopia_code_ = top_code_;
+      second_diplopia_code_ = second_code_;
+    }
+  }
   top_n_flags_[null_char_] = TN_TOP2;
 }
 
@@ -1138,6 +1174,10 @@ void RecodeBeamSearch::PushHeapIfBetter(int max_size, int code, int unichar_id,
     if (UpdateHeapIfMatched(&node, heap)) {
       return;
     }
+    // check to see if node is possible diplopia
+    if (!AddToHeapIsAllowed(&node)) {
+      return;
+    }
     RecodePair entry(score, node);
     heap->Push(&entry);
     ASSERT_HOST(entry.data().dawgs == nullptr);
@@ -1189,6 +1229,20 @@ bool RecodeBeamSearch::UpdateHeapIfMatched(RecodeNode *new_node, RecodeHeap *hea
   return false;
 }
 
+// Determines if node can be added to heap based on possible diplopia status
+bool RecodeBeamSearch::AddToHeapIsAllowed(RecodeNode *new_node) {
+  if (!in_possible_diplopia_)
+    return true;
+  const RecodeNode *prev_node = new_node->prev;
+  if (prev_node != nullptr && prev_node->code == first_diplopia_code_ && new_node->code == second_diplopia_code_) {
+    return false;
+  }
+  if (prev_node != nullptr && prev_node->code == second_diplopia_code_ && new_node->code == first_diplopia_code_) {
+    return false;
+  }
+  return true;
+}
+
 // Computes and returns the code-hash for the given code and prev.
 uint64_t RecodeBeamSearch::ComputeCodeHash(int code, bool dup, const RecodeNode *prev) const {
   uint64_t hash = prev == nullptr ? 0 : prev->code_hash;

src/lstm/recodebeam.h

@@ -377,6 +377,9 @@ class TESS_API RecodeBeamSearch {
   // Searches the heap for an entry matching new_node, and updates the entry
   // with reshuffle if needed. Returns true if there was a match.
   bool UpdateHeapIfMatched(RecodeNode *new_node, RecodeHeap *heap);
+  // Determines if new node can be added to the heap for the current beam.
+  // Returns false if we are in possible diplopia situation
+  bool AddToHeapIsAllowed(RecodeNode *new_node);
   // Computes and returns the code-hash for the given code and prev.
   uint64_t ComputeCodeHash(int code, bool dup, const RecodeNode *prev) const;
   // Backtracks to extract the best path through the lattice that was built
@@ -425,6 +428,11 @@ class TESS_API RecodeBeamSearch {
   bool is_simple_text_;
   // The encoded (class label) of the null/reject character.
   int null_char_;
+  // Variables used in tracking possible diplopia case
+  // Refer to ComputeTopN routine for use of these variables
+  bool in_possible_diplopia_;
+  int first_diplopia_code_;
+  int second_diplopia_code_;
 };
 
 } // namespace tesseract.