multi task training discuss

[carry-xz]

nice work！
What kind of tasks are suitable for training with a multitasking framework? Suppose that the data of different tasks are different, such as the two tasks of vehicle attributes and vehicle quality, how to unify the training with different data? Vehicle quality data may contain blurred/obscured/non-vehicle images, making it difficult to achieve consistent labeling. Just to discuss