The caption just looks strange.

[LibraSkywalker]

I'm just using such model to caption my pictures and I found two problems.
The first one is, it'll shuffle my pictures. I don't know why but such thing do happen. I can see it in the log:

cp "/home/libraskywalker/Project/neuraltalk2/image/91.jpg" vis/imgs/img1.jpg
image 1: a kitchen with a table and chairs and a stove
evaluating performance... 1/-1 (0.000000)
cp "/home/libraskywalker/Project/neuraltalk2/image/181.jpg" vis/imgs/img2.jpg
image 2: a cell phone sitting on top of a table
evaluating performance... 2/-1 (0.000000)
cp "/home/libraskywalker/Project/neuraltalk2/image/189.jpg" vis/imgs/img3.jpg
image 3: a bowl of fruit is on a table
evaluating performance... 3/-1 (0.000000)
cp "/home/libraskywalker/Project/neuraltalk2/image/5.jpg" vis/imgs/img4.jpg
image 4: a black and white cat sitting on top of a white cat
evaluating performance... 4/-1 (0.000000)
cp "/home/libraskywalker/Project/neuraltalk2/image/108.jpg" vis/imgs/img5.jpg
image 5: a vase with a flower in it on a table
evaluating performance... 5/-1 (0.000000)
cp "/home/libraskywalker/Project/neuraltalk2/image/52.jpg" vis/imgs/img6.jpg
image 6: a car is parked on the beach near the water
evaluating performance... 6/-1 (0.000000)
...

And also the caption it self is some how really weird, for example the picture is about a bunch of books on the book shelf. However, the caption is "a pair of scissors on a wooden table". I don't understand why the book will be labeled as the scissors?