MiracleDance / PoseRAC

PoseRAC: Pose Saliency Transformer for Repetitive Action Counting
MIT License
13 stars 2 forks source link

hard sample mining #20

Open gezi1103 opened 6 months ago

gezi1103 commented 6 months ago

When selecting hard samples, the input labels are obtained based on the index of the maximum value in each row of y. For example, for the prominent pose 1 of front_raise, the label is [10000000], and its maximum value index is 0. For the prominent pose 2 of front_raise, the label is [00000000], and the maximum value index is also 0. This would make them considered matching samples. How are the two distinguished?