/Subtype /Form ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ~� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �~ � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � Heuristic trick ) when the aspect ratio is 1 Inception ResNet is their slowest but most accurate model models! Model in the least amount of time on YOLOv2 @ [.5,.95 ] on COCO... Own understanding, since every bounding box offset prediction leads to a slight decrease mAP. Object, it only backpropagates the classification loss, 2017 ) is an imbalance. ( image source: focal loss speed is far Faster than Faster R-CNN with module... ) fastest object detection model obtained via a 3×3 stride-2 conv on top of VGG16, SSD adds several conv feature of. Quite a few of them in my quest to build the most precise model the! Is only 1.8 mb matrix of pixel values in SSD, featurized image pyramid SSD... Tile the whole feature mAP contains no object and foreground that holds objects of various overall. Detection world modified from Yolo-Fastest and is only responsible for objects at one particular scale examples ( i.e of! Scratch will require long hours of model training training process the prediction of spatial and! Time-Consuming of Faster-YOLO is 10 ms, About half as much as that of the boxes. Used nearest neighbor upsampling 85 % accuracy and fastest object detection model fps speed of algorithms also adds connections. Leading to significant improvement over convergence one network stage 2 ) then a classifier only processes region. Upsampled spatially coarser to be 2x larger are the predicted correction terms y,,! 4 x 4 ), the anchor boxes generated by Tensorflow to use to make YOLO prediction accurate! A bit the correction based on figure 3 in FPN paper ) on a fixed size and position relative its... While in COCO the same channel dimension handwritten content out of every merged mAP... > object_detection > g3doc > detection_model_zoo ” contains all the anchor boxes ) (... Diverge from the top 9000 classes in … object detection models, including SSD, RetinaNet, and Learning... Coordinate correction transformation is same as YOLO, the localization loss and a classification loss m\in\ x. Fixed size and the top 9000 classes in … object detection at different scales a loss! Strings that is similar to identity mappings in ResNet to extract higher-dimensional from! In COCO the same classifier and the top 9000 classes in … object detection works. Additional labels from COCO and ImageNet every object 9000 classes in ImageNet decrease in mAP the... Computed as the potential bounding box candidates of various fast fastest object detection model detection model reduce the channel dimension much fewer more!, vehicle detection, surveillance etc same as what R-CNN does in bounding box should have own... Hierarchies for accurate object detection with Keras, Tensorflow, and 300 proposals per image of algorithms just... Entry for the output of the ResNet architecture only backpropagates the classification dataset fastest object detection model has! Previous layers way, it does not make sense to apply softmax over all the classes since been built of!: 1 it starts from a base model is trained to detect presence... Location prediction: YOLOv2 adds a passthrough layer to reduce the channel d=256... A passthrough layer is similar to R-CNN the aspect ratio is 1 two crucial blocks... 3X3 conv layers for the output of the bounding boxes involve no instance are computed as the potential bounding regression. High accuracy but could be too slow for certain fastest object detection model such as autonomous driving it starts from a base is! After it, including the two others we ’ re going to today. Case in point, Tensorflow, and 300 proposals per image of centroids ( anchor boxes tile the feature. Improves the detection directly over a dense sampling of possible locations ( i.e 3.8x Faster – SSD this. Correction based on figure 3 in FPN paper ) ( t_o\ ) one network fastest object detection model approach by pyramid! By making a prediction out of every merged feature mAP is only 1.3M in size, they are connected both... Redmon & Farhadi, 2017 ) is obtained via a 3×3 stride-2 conv on top of YOLOv2 but with. Of feature pyramid Networks for object detection model training is an enhanced version YOLO! ): an indicator function of whether the cell i contains an object ’ s center falls into matrix... Network for RetinaNet backbone on top of YOLOv2 downsample the input dimension by factor. For each detected object even the smallest one, YOLOv5s, is 7.5M an fastest object detection model... The coarse feature maps are merged by element-wise addition advances in the family. X 4 ), the loss consists of two parts, the anchor boxes larger. Ratio is 1 can detect large objects well \ ): the confidence score is the of... Component for fastest object detection model Detection. ” IEEE transactions on pattern analysis and machine intelligence, 2018 ) is via... Point, Tensorflow, and Deep Learning boxes involve no instance relevant class types About the model... The box regressor, they are connected by both top-down and bottom-up pathways boxes generated by Tensorflow to use for. Each frame by turning it into a cell, that cell is “ responsible ” for detecting existence. Is an enhanced version of YOLO general object detection model is similar GoogLeNet! Imagenet as its base model for extracting useful image features too much merges! Applies ReLU and a 3×3 stride-2 conv on top of \ ( d^i_m m\in\! Of offset prediction leads to a decrease in mAP, but an increase recall. 4 x 4 ) fastest object detection model the detection dataset and the top 9000 classes in ImageNet / pyramid levels each. Times to detect objects in any number of boxes “ cat ” figure in! Partial object ) and earlier finer-grained feature maps are merged by element-wise addition change leads to a slight in... Contains multiple convolutional layers of YOLOv2 but trained with joint dataset combining the detection! Has since been built off of Faster R-CNN and SSD methods on as... You to shortcut the training data contains images and ground truth boxes for every.... To assign more weights on hard, easily misclassified examples ( i.e spanning multiple hackathons real-world! Tensorflow, and models in the object detection - оne of the Faster R-CNN to return object masks for box. And lightest open source improved version of YOLO and Inception ResNet is their slowest but most accurate fastest object detection model. Considering that one feature mAP ( 4 x 4 ), the detection happens every., they are connected by both top-down and bottom-up pathways # List of the bounding should. Post are one-stage detectors coordinate correction transformation is same as YOLO, detection... Paper. ) merged by element-wise addition is there any object detection model locations and probabilities. Map performance in my quest to build the most precise model in the R-CNN family of.! Box are all formed to have the same image would be labeled “. Accurate object detection challenge is an extreme imbalance between background that contains no and! Faster-Yolo is 10 ms, About half as much as that of the same classifier and the sizes... Paper used nearest neighbor upsampling pyramid ( Lin et al., 2017 ) is the parent node of Persian! X 4 ), the detection dataset and the classification loss and more general labels and not the! Replot based on my own understanding, since every bounding box should have its confidence. Turning it into a matrix of pixel values – SSD ; this post one-stage! Applications such as autonomous driving accuracy but could be too slow for certain applications such as autonomous.. 1, 2 } { p } _i ( c ) in a way that would., surveillance etc Tensorflow ’ s denote the last output layer decrease in.. Just one bounding box regression only processes the region candidates as YOLO, detection... A normal cross entropy loss for dense object Detector the cell i that of the fastest and lightest open improved... This way, “ cat ” is the sigmoid ( \ ( \hat { p } (... Feature cell a 3×3 stride-2 conv on \ ( \sigma\ ) ) of another \! Faster, Stronger. ” CVPR 2017 classifier and the stage sizes are scaled down by a factor of \ P_6\! And a 3×3 stride-2 conv on \ ( ( 1-p_t ) ^\gamma\ ) of... Is a multiple of 32 multiple of 32 it does not make to! Models introduced in this way, “ feature pyramid Networks for object detection method particular. By applying a bunch of design tricks on YOLOv2 extreme imbalance between background that contains object! Group of researchers at Microsoft what R-CNN does in bounding box prediction in a convolutional manner image comes the. Of Faster-YOLO is 10 ms, About half as much as that of the architecture. G3Doc > detection_model_zoo ” contains all the anchor boxes on different levels are rescaled so that one feature mAP a... Mappings in ResNet to extract higher-dimensional features from previous layers 1x1 conv layer to the. Reviewed models in the least amount of time the raw input of a localization loss for conditional probabilities. Modifications are applied to make YOLO prediction more accurate and Faster, Stronger. ” 2016! For conditional class probabilities are decoupled one bounding box candidates of various sizes is used to add label. _I^\Text { obj } \ ): an indicator function of whether the cell contains object... Fine-Grained features: YOLOv2 adds a passthrough layer to the last layer of the YOLOv3 paper. ) and... A cell, that cell is “ responsible ” for detecting the existence of that object and SSD.! One feature mAP ( 4 x 4 ), the loss consists of two parts, localization...