Revisiting Referring Expression Comprehension Evaluation in the Era of Large Multimodal Models.