VQA-Real [5] Q: What shape is the bench seat? Q: What color is the stripe on the train? Q: Where are the magazines in this picture? A: Curved A: White A: On Stool Q: What color is the clock? A: Green Q: What is the woman doing? A: Sitting Q: How is the ground? A: Dry Q: Who looks happier? A: Man Q: Where are the flowers? A: Around Tree Q: How many pillows? A: 2 Q: What color are the barricades ? A: Pink Q: What are they using to draw ? A: Markers Q: Who is playing? A: Rafael Nadal Visual Genome [35] VQA-Abstract [5] Zero-Shot VQA [70] VQA v2.0 [25] Q: Who is wearing glasses? A: Man A: Woman Q: Is the TV on? A: Yes A: No FIGURE 2. Examples from the test splits of different VQA data sets. For the zero-shot VQA data set, the highlighted words are unknown words, i.e., not present in training examples. 66 IEEE SIGNAL PROCESSING MAGAZINE | November 2017 |