Skip to content
The methodOpen data
Back to the collection

Predictive Inequity in Object Detection

Benjamin Wilson, Judy Hoffman, Jamie Morgenstern

2019 · arXiv

Computer visionPsychology
Read the paper
HN score
0
Mentions
4
Readers
4
Months
3
What’s in a score?

From the abstract

In this work, we investigate whether state-of-the-art object detection systems have equitable predictive performance on pedestrians with different skin tones. This work is motivated by many recent examples of ML and vision systems displaying higher error rates for certain demographic groups than others. We annotate an existing large scale dataset which contains pedestrians, BDD100K, with Fitzpatrick skin tones in ranges [1-3] or [4-6]. We then provide an in-depth comparative analysis of performance between these two skin tone groupings, finding that neither time of day nor occlusion explain this behavior, suggesting this disparity is not merely the result of pedestrians in the 4-6 range appearing in more difficult scenes for detection. We investigate to what extent time of day, occlusion, and reweighting the supervised loss during training affect this predictive bias.

Metadata: arXiv. Abstracts retain their original authors’ rights.

In the conversation

Original HN comments, with labels from simple text rules. A label describes the whole comment and may miss context or irony. Read the discussion to make up your own mind. About the labels

4 selected excerpts · 4 total mentions

js2
Neutral mention · rule
The story doesn’t cite the specific paper. But I think I submitted the paper last month: “Predictive Inequality in Object Detection” https://news.ycombinator.com/item?id=19379827 https://arxiv.org/abs/1902.11097 I’m not denying the conclusion of the paper, but it does have a lot of limitations. Also, Tesla notwithstanding, isn’t object detection primarily done via lidar?
Original comment MIT Researcher Exposing Bias in Facial Recognition Tech Triggers Amazon’s Wrath
uberman
Neutral mention · rule
The article is very thin on details. It serves mostly to point to an abstract for a research paper : https://arxiv.org/abs/1902.11097 The abstract is unfortunately also very light on detail. It appears as though the researchers trained a number of classification systems using static images of pedestrians noting skin color and observed that darker skin tones resulted in mis-classification more often than for lighter skin tones. The original article then sensationalizes this suggesting that means self driving cars are more likely to kill darker skinned people. I believe there are a number of problems here. Fist, the skin tone bias in image training data sets is well…
Read the full comment
documented. The abstract it self says such bias motivated their research. The conclusion that pedestrians with darker skin tones were mis-classified more often should not really come as a surprise given that these training sets have historically resulted in this type of mis-classification. Second, pedestrian detection is not done (typically) using cameras but with LIDAR. LIDAR has no notion of skin tone. Cameras are more traditionally used for lane departure and blind spot checking. See : https://www.quora.com/How-do-self-driving-cars-see-in-the-da... In my opinion there there is quite a large gap between noting the impact of biases in training data applied to static images of pedestrians and a sensationalized conclusion that "Self-driving cars may be more likely to hit you if you have dark skin"
Original comment Self-driving cars may be more likely to hit you if you have dark skin
czr
Neutral mention · rule
The headline here is not a good description of the actual paper (https://arxiv.org/pdf/1902.11097.pdf). The difference in average precision in bounding boxes between the LS (light skin) and DS (dark skin) observed by the researchers was around 1-10%. Not "can't recognize". Notably, people with dark skin were underrepresented in the dataset that the researchers used. When they compensated for the dataset imbalance by reweighting examples, the issue was resolved. Notably, the research itself is on publicly-available object detection systems, not any actual autonomous vehicles. Any company seriously building an AV 1) has their own diverse datasets collected from real-world…
Read the full comment
driving and 2) has almost certainly already handled this issue (skin color is a blessedly minor source of variation in pedestrian detection, compared to the tremendous headaches caused by clothing, costumes, carrying other objects, all of which need to be accounted for...). The paper is good a reminder of the importance of fine-tuning any pedestrian detection system to fix imbalances in your dataset, and avoid false negatives. But it's not what the title claims.
Original comment Autonomous Cars Can't Recognise Pedestrians with Darker Skin Tones

Keep following the thread.

More in Computer vision

More from the same topics.

2018arXiv

Talent vs Luck: the role of randomness in success and failure

A. Pluchino. A. E. Biondo, A. Rapisarda

The largely dominant meritocratic paradigm of highly competitive Western cultures is rooted on the belief that success is due mainly, if not exclusively, to personal qualities such as talent, intelligence, skills, efforts or risk taking.

Psychology
16 mentions14 months
2.5HN score