the performance is so bad

[junchen14]

Hi when I tried evaluation with neural baby talk, it seems the performance is very bad! is this normal? for example. image 384213: wipe jetliner doing wipe chickadee payer wipe nuts image 386164: wipe moped tried operate doing at rubs moped persons image 223648: wipe moped ball operate doing at rubs moped jello image 403385: wipe shamrock doing wipe frisbee chickadee payer sled image 294832: wipe shamrock doing wipe chickadee frisbee payer sled image 462565: wipe songbird undergoing wipe waterfowl topless wipe baseline bathtub useful wipe songbird image 436141: wipe shamrock doing wipe chickadee frisbee payer preparing image 192440: wipe shamrock doing wipe frisbee payer wipe chickadee image 1146: wipe songbird corsage wipe cactus payer ostrich doing commode image 559665: wipe songbird undergoing wipe bluejay topless wipe baseline image 394240: wipe bluejay grounds corsage dandelion rubs wipe mates image 491497: wipe gentleman skating outdoor heating rubs wipe afraid canoe image 184791: wipe mommy rubs wipe mommy rubs wipe doggie rubs bathes

the sentence is not very consistent at all