Using scraped data from Airbnb's platform in New York City alongside state-of-the-art Vision Transformer models for image classification, this paper investigates the existence and extent of race discrimination in the Airbnb platform and the impact of a policy to reduce this bias. First, we show that Black hosts have a 7.2 percentage points lower occupancy rate than their White counterparts despite no differences in pricing. For Asian and Hispanic hosts, the difference compared to White hosts is small and largely insignificant for both occupancy rates and prices. Second, using difference-in-differences and event-study approaches, we show that the 2018 Airbnb anti-discrimination policy, which reduced the size of users' profile pictures on the platform, unexpectedly increased the Black-White disparity by about 3.3 percentage points. As a reaction to the adverse impact of the new policy, Black hosts increased the number of amenities in their listings. A potential mechanism for the increase in Black-White disparity stems from the increasing guests' uncertainty in discerning facial features that positively correlate with occupancy rates from the smaller profile pictures. As a result, guests focus more on skin color.