Open CrazyLenmon opened 9 months ago
In uni-perceiver V2, Sk denote the sampling ratio of each task. When using unmixed sampling, you only sample one task per iteration, so does the value of K in LOSS equal to 1 ? If that so, does Sk make a real difference?
In uni-perceiver V2, Sk denote the sampling ratio of each task. When using unmixed sampling, you only sample one task per iteration, so does the value of K in LOSS equal to 1 ? If that so, does Sk make a real difference?