Model	Human Metric	Auto Metric	Identify (Binary Accuracy)
Humans	95		92
Ground-truth Caption → Llama-2-7b (Oracle)	71
Ground-truth Caption → Llama-2-13b (Oracle)	70
Ground-truth Caption → GPT4 (Oracle)	69
Ground-truth Caption → GPT3 (Oracle)	68	62	74
BLIP2 FlanT5-XXL (Fine-tuned)	27	24	73
BLIP2 FlanT5-XL (Fine-tuned)	15	18	60
Predicted Caption → Llama-2-7b	36
Predicted Caption → Llama-2-13b	36
Predicted Caption → GPT4	36
Predicted Caption → GPT3	33	42	59
InstructBLIP	    31
LLaVA	    31
mPLUG-Owl       24
BLIP2 FlanT5-XXL (Zero-shot)	0	0	50