Analyze the failures of a vision model

A vision model works well on average but fails on some inputs. Explain how you would find and group the failures (slicing by condition, looking at examples, confusion analysis), tell data problems from model problems, and decide what to do next: more or better data, augmentation, a different model, or changed thresholds.