how to train this model?
carry-xz opened this issue · comments
nice work! Is it re-annotates the Wider Face dataset and retrains the three branches model ?
- 采用hopenet预训练模型,在widerface的基础上,为每个人类作姿态估计label标注
- 借鉴refer口罩分支检测的思路,没有作loss上面的改动,只是单纯的在数据上合并,加mosaic,网络分支修改等简单操作罢了。