人像在b2分支上做SOC

[buptlihang]

在新的任务上SOC一般来说需要修改以下几个配置项

• 修改数据集路径config.sample_path
• 修改分类数config.cls_num
• 修改类别权重config.classes_weight
• 修改mean, std config.mean和config.std

训练时train.py可修改项

···
backup_detail_loss = boundaries * F.cross_entropy(pred_detail, backup_detail.max(1)[1], weight=self.config.classes_weight, reduction='none')
···
backup_fusion_loss = boundaries * F.cross_entropy(pred_fusion, backup_fusion.max(1)[1], weight=self.config.classes_weight, reduction='none')
···
self.config.loss = 5*soc_semantic_loss + backup_detail_loss + backup_fusion_loss

两个loss的weight参数可有可无，也可随机组合。soc_semantic_loss 前的系数可改动