M.S. AAI Capstone Chronicles 2024

also differences in the level of detail of each caption and the focus of the caption (e.g., only one caption mentions the color of the clothing, one caption does not mention the flowers). The captions are about 12 words in length on average, but range from as short as 2 words to as long as 78 words. Figure 2 shows this distribution as a box plot. 95% of the captions were 22 words or less, with a substantial number of outliers much larger than this. This is significant because for a model that processes text sequences, the sequences need to be uniform in length. Figure 2 Distribution of Image Caption Lengths

​ To get a better idea of what subjects appear most commonly in the images, we used a

word cloud to visualize the text corpus (excluding stop words such as “a” or “and”). The size of the words indicates how frequently they appear in the corpus. In the word cloud shown in Figure 3 on the next page, “man” and “woman” are by far the most prominent words, suggesting that most of the images in the dataset are of people.

209