Yes, multiple studies and real-world cases have shown that many facial recognition systems perform poorly on some races (especially non-white groups) in terms of accuracy. This phenomenon is usually caused by the following reasons: 1. Data imbalance: Many facial recognition algorithms are trained on datasets that are dominated by white faces, resulting in a lack of sufficient samples and diversity when identifying other races, resulting in poor performance. 2. Feature extraction bias: Due to the subtle differences in facial features between different races, the algorithm may not be optimized for all groups when extracting and matching features, resulting in increased error rates. 3. Algorithm design issues: Some algorithms may not fully consider the differences between different races when designing, resulting in reduced recognition accuracy for some races in actual applications. 4. Insufficient labeling and verification: If there are not enough samples of different races for verification during training and testing, bias errors are prone to occur. In recent years, the industry has paid more and more attention to this issue, promoting improvements in dataset diversity, algorithm fairness, and model tuning. For example, researchers will use richer and more inclusive training data, or design specialized model architectures to improve performance on various races. In summary, although facial recognition technology continues to improve, there are still challenges in the accuracy of recognition of certain races, and further optimization and supervision are needed to ensure fairness and reliability.