mmdetection使用coco预训练

mmdetection中configs文件一般默认是使用imagenet预训练的backbone权重参数，但是使用coco预训练一般来说会使模型收敛更快，效果更好，是比赛提分的一个小trick！
1.修改coco预训练模型的类别不一致问题

import torch
num_classes = 21
model_coco = torch.load("cascade_rcnn_x101_32x4d_fpn_2x_20181218-28f73c4c.pth")
    
# weight
model_coco["state_dict"]["bbox_head.0.fc_cls.weight"] = model_coco["state_dict"]["bbox_head.0.fc_cls.weight"][:num_classes, :]
model_coco["state_dict"]["bbox_head.1.fc_cls.weight"] = model_coco["state_dict"]["bbox_head.1.fc_cls.weight"][:num_classes, :]
model_coco["state_dict"]["bbox_head.2.fc_cls.weight"] = model_coco["state_dict"]["bbox_head.2.fc_cls.weight"][:num_classes, :]
# bias
model_coco["state_dict"]["bbox_head.0.fc_cls.bias"] = model_coco["state_dict"]["bbox_head.0.fc_cls.bias"][:num_classes]
model_coco["state_dict"]["bbox_head.1.fc_cls.bias"] = model_coco["state_dict"]["bbox_head.1.fc_cls.bias"][:num_classes]
model_coco["state_dict"]["bbox_head.2.fc_cls.bias"] = model_coco["state_dict"]["bbox_head.2.fc_cls.bias"][:num_classes]
#save new model
torch.save(model_coco,"coco_pretrained_weights_classes_%d.pth"%num_classes)

2.configs文件设置

load_from = './models/coco_pretrained_weights_classes_21.pth'
#load_from = None
resume_from = None

mmdet默认加载权重优先级别是resume_from(断点加载)，load_from，pretrained的顺序，所以需要从load_from加载预训练权重