search

peteryuX / retinaface-tf2

RetinaFace (Single-stage Dense Face Localisation in the Wild, 2019) implemented (ResNet50, MobileNetV2 trained on single GPU) in Tensorflow 2.0+. This is an unofficial implementation. With Colab.
MIT License
272 stars 97 forks source link

classification loss (crossentropy) #23

Open onlyNata opened 4 years ago

onlyNata commented 4 years ago 
    # classification loss (crossentropy)
    # 1. compute max conf across batch for hard negative mining
    loss_class = tf.where(mask_neg,
                          1 - class_pred[:, 0][..., tf.newaxis], 0)

請問，這裏的 分類損失 為什麽是這樣的計算的？要是檢測的目標不僅僅是人臉，而是多個目標（比如，人臉和行人等）是否要做相應的修改？