/w=32,quality=90,fit=scale-down){kind=small}
Raphael Thys
An analysis of gender representation in Disney and Pixar films using computer vision models I trained
by author
I’ve always loved animated films. As a child, I would watch my favorite Disney films repeatedly on my family’s VCR. This was in the days before DVDs, decades before streaming. As an adult, I haven’t grown out of it (there’s nothing to grow out of). In fact, whenever there is a new animated movie in cinemas, I like to treat myself to a viewing.
When I compare the films released when I was a kid with the ones produced nowadays, I feel like there are important differences in how stories are told and characters are depicted. It feels like there is more diversity of representation.
A few months ago, I decided to find out if that really was the case by using neural networks to measure the equality of gender representation in animated films. While all forms of representation matter (gender, ethnic, social-economic, LGBTQ+…), gender in the traditional binary sense (male and female) is easier to measure, so I started there.
Here’s what I found:
- The percentage of female faces in animated films is on average lower than the percentage of male faces, although there is an overall positive trend towards more equal gender representation. (Read more)
- Gender representation is not becoming more equal uniformly across animated films, but rather movies with female main characters are becoming biased towards female faces to the same extent that films with male main characters are biased towards male faces. (Read more)
- When characters appear in a group, the male characters tend to outnumber the female characters. In fact, groups with more than one woman are rare. (Read more)
- There are films where the female protagonist exists in a male-dominated world, but none yet where the male protagonist is in a female-dominated world. (Read more)
- There is a correlation between equality of representation and fairness of representation. (Read more)
Check out the computer vision models I trained on this demo website.
The Rise in Female Representation
Using two neural networks, one trained to detect animated faces and one trained to classify the gender of animated faces, I analyzed feature-length Disney and Pixar films (1937 — 2021) with a human or humanoid cast of characters. The models counted the number of male and female faces in each film to calculate the male-to-female face ratio.
Workflow: First, the face is localised (yellow) by the face detector, then the gender is determined (pink/ blue) by the classifier. My face detector and gender classifier can be interacted with on this demo website.
While the percentage of female faces in relation to male faces has reached new heights in recent years, many films continue to have few female faces overall. In fact, the film with the second lowest proportion of female faces, Onward (2020), was released only three years ago.
Percentage of female (pink) vs males (light blue) faces per film
Grouping the films by decade show a clearer positive trend towards equal gender representation, though a slow one. In this decade, female characters are still underrepresented at a ratio of three men for every two women.
Grouping the films also reveals a higher standard deviation in recent decades, meaning that the percentage of male and female faces in a film oscillates more for newer productions. This suggests greater diversity — some films are heavily male while women dominate others. In contrast, the smaller standard deviation in older films could suggest that those movies were more formulaic, with similar structures and characters.
Average percentage of female (pink) vs male (light blue) faces per decade. White numbers inside columns indicate how many films were grouped per decade, and grey error bars and numbers indicate standard deviation from the decade mean. The dashed red line indicates the linear fit through the data, with ‘x’ representing the number of decades after 1930.
The Main-Character Effect
By segmenting films based on the main character’s gender, more nuanced insights can be reached. Whereas films with female protagonists show an increasing proportion of female faces, that is to say, an increase in the “female face ratio” (FFR), films with male protagonists don’t.
In other words, gender representation has not become more equal uniformly across films, but rather films with female main characters are doing the heavy lifting. And by steadily increasing their percentage of female faces, they have created a positive overall trend.
Percentage of female faces (FFR) per film segmented by the gender of the main character.
Percentage of female faces (FFR) in films where the main character is female.
In fact, they have increased their FFR so much that some recent movies with female main characters are biased towards female faces (e.g. Inside Out (2015), Raya and the Last Dragon (2021), Encanto (2021)).
This phenomenon explains the small number of gender-balanced films, despite movies increasing their FFR over the years. Out of 51 analyzed films, only seven met the criteria of showing female faces 45% — 55% of the time.
Female Presence in Groups
By counting the amount of male and female faces present in a frame, the different character combinations in a film can be determined. This form of analysis revealed that movies with male protagonists emphasize male-dominant groupings.
Not only are there many all-men groups, but the 1-male and 2-male combinations are also the two most common in films with male main characters, resulting in no women whatsoever being seen in over 55% of frames. There is also a big gap between the 1-male and 1-female cases, which isn’t mirrored by movies with female main characters.
Male (m) and female (f) character combination in films with male (blue) and female (pink) protagonists.
Temporal analysis shows that frames containing no female faces are the most common and have remained so for almost as long as animated films have existed.
Frames with three or more females are rare and have only begun appearing in the last two decades.
Average number of female faces per frame by decade.
Female Prominence
So far, the presence of faces has been analyzed, but the prominence of faces can also be studied by measuring face sizes. Prominence is a good indicator of the importance assigned to a character —whereas a small face could belong to a secondary character in a crowd, a big face will usually belong to a character that is central to the story.
Large faces are, by definition, faces with bounding boxes greater than 96² pixels. An analysis of all film frames showed that over 66% of faces in Disney and Pixar movies were large, so a strong correlation between the percentage of large female faces and the percentage of all female faces was to be expected:
Correlation between the percentage of large female faces and all female faces. The black line marks a 1:1 correlation, and the red line shows a linear fit through the data.
Since the linear correlation is strong, outliers are all the more interesting. The film Brave (2012) is one such outlier, with its female faces being 10.8% more prominent than male faces.
In the case of Brave (2012), the high female face prominence can be attributed to the plot, which revolves around a girl, Merida, seeking emancipation in a male-dominated world (a deeply patriarchal, medieval community). Curiously, no opposing outlier represents a story about a male main character living in a female-dominated world. Disney, Pixar, if you’re reading this, maybe that’s a new storytelling opportunity!
Difference between the percentage of female faces overall (FFR) and large female faces (large-face FFR).
In general, it was found that films with a female main character have a large-face FFR above the overall FFR. In fact, since the 1990s, large-face FFR has always been higher than the baseline for movies with female protagonists (there are no pink bars with negative values after 1991).
It is tempting to explain this away by saying that more attention is always given to the film’s protagonist, so high female face prominence is to be expected in films with female main characters.
The problem with this logic is that films with male main characters don’t follow the rule — male films are scattered more or less equally around the FFR baseline, meaning that equal prominence is given to male and female faces.
Then again, there is an underlying bias in animation in favor of male faces, so it could be that the prominence given to the main character in male-protagonist films doesn’t register because they are already in a male-dominated world.
Equal vs Fair Representation
It is worth pointing out that, so far, equality of representation has been measured. This is different from the fairness of representation. For example, a film that has an equal number of male and female characters, and gives them equal screentime, can still represent gender unfairly if the characters behave in stereotypical, gender-normative ways.
To examine if animation films represent gender fairly, each film’s Bechdel Test score, as recorded in the bechdel test database, was checked. The Bechdel Test was invented by a cartoonist as a joking way to measure the fairness of female representation.
Funnily enough, it’s been widely adopted as an official measure. To pass the test, a film must have at least two women in it, who talk to each other about something besides a man.
Bechdel Test scores of Disney and Pixar animated movies over the last three half-centuries. White numbers in columns indicate how many films were grouped per time period. Error bars indicate the standard deviation from the FFR mean.
Although the Bechdel Test sets a pretty low bar, only 25% of films passed the test in the first half of the 20th century. Fortunately, since the turn of the century, over 78% of films have passed the test. This development shows a positive correlation between fairness of female representation and equality of female representation, with the animation industry being faster in their conversion to fair female representation than equal representation.
Conclusion
To summarise, female representation in animated films is still lagging behind male representation on average, but there are steady signs of improvement. Both equality and fairness of gender representation are increasing, and there are many films where the percentage of female faces even exceeds that of male faces.
Storytelling is an art form, and it would be limiting to prescribe a “correct” amount of female or male screen time. Different stories are just that: different. Some might require a greater or smaller number of female characters, and those characters might be more or less important to the plot. With that said, it is good to keep in mind that the world is made up of men and women and that sometimes a male character could just as easily be rewritten as a female character.
Without being too prescriptive, it is important for gender to be well represented because films can convey subtle messages that might influence how audiences perceive others and even themselves. If animated movies with male protagonists are primarily viewed by boys and those with female protagonists are primarily viewed by girls — and in both cases, the films regularly have gender imbalances — couldn’t that leave unintended impressions on young kids about the importance of the opposite sex or the spaces they occupy in society? And if boys or girls recurringly see characters they identify with in specific roles or are given a certain level of prominence, couldn’t that shape their behavior and ambitions to some extent?
By letting children (and adults) see their gender represented in a wide variety of roles, films can encourage viewers to imagine themselves in those roles, dissuade them from setting artificial boundaries for themselves and inspire them to live up to their full potential.
Acknowledgments
- Thank you to Animation Screencaps for making the film frames of so many animated films accessible for non-commercial use.
- Thank you to the Detectron2 library creators and contributors for making state-of-the-art computer vision models freely available for training.