∵0**0=1 ∴ gamma != 0

[zhangyuqi-1]

增加 gamma = 0 的情况的讨论

def focal_loss_sigmoid(prediction_tensor, target_tensor, weights=None, alpha=0.5, gamma=0):
    target_tensor = tf.cast(target_tensor, tf.float32)
    sigmoid_p = tf.nn.sigmoid(prediction_tensor)
    zeros = array_ops.zeros_like(sigmoid_p, dtype=sigmoid_p.dtype)

    pos_p_sub = array_ops.where(target_tensor > zeros, target_tensor - sigmoid_p, zeros)
    neg_p_sub = array_ops.where(target_tensor > zeros, zeros, sigmoid_p)

    if gamma != 0:
        per_entry_cross_ent = - alpha * (pos_p_sub ** gamma) * tf.log(tf.clip_by_value(sigmoid_p, 1e-8, 1.0)) \
                              - (1 - alpha) * (neg_p_sub ** gamma) * tf.log(tf.clip_by_value(1.0 - sigmoid_p, 1e-8, 1.0))
    else:
        # ∵0**0=1 
        per_entry_cross_ent = - alpha * target_tensor * tf.log(tf.clip_by_value(sigmoid_p, 1e-8, 1.0)) \
                              - (1 - alpha) * (1 - target_tensor) * tf.log(tf.clip_by_value(1.0 - sigmoid_p, 1e-8, 1.0))

    return per_entry_cross_ent

[zhangyuqi-1]

此即为sigmoid_cross_entropy

[marvinquiet]

Agreed. I also found that when gamma=0, it does not correctly compute the loss.