prediction:
tensor([5.3048e-04, 9.0192e-04, 6.0176e-04, 1.1318e-04, 6.6526e-03, 1.0745e-03,
        1.3762e-03, 9.9780e-01, 8.6294e-03, 3.1465e-04, 4.0600e-03, 3.5634e-03,
        3.5838e-04, 8.7827e-04, 3.8405e-02, 3.0164e-03, 3.6145e-04, 8.4642e-03,
        3.0172e-03, 7.4142e-03])
        
gt:
tensor([0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0.])

mAP(pred=prediction_batch, target=labels_batch)

将 tensor 按照某个阈值而到 1 和 0

prediction_threshold = torch.where(prediction_batch > 0.5, 1, 0)