地址链接:https://www.flyai.com/m/resnext101_32x4d_gn-ac3bb84e.pth
查看来源:https://download.openmmlab.com/pretrain/third_party/resnext101_32x4d_gn-ac3bb84e.pth
确定自己使用的框架并导入对应的库。导入库实现样例代码可参考 文档中心-预训练模型使用教程
在代码中实现加载预训练模型地址
import torchvisionfrom flyai.utils import remote_helper
path=remote_helper.get_remote_data("https://www.flyai.com/m/resnext101_32x4d_gn-ac3bb84e.pth")
model = torchvision.models.res2net101(pretrained = False) # model = torchvision.models.res2net101(pretrained = True) # 这行代码与上面等同,只不过一个是调用FlyAI提供的预训练模型地址,一个是外网的地址model.load_state_dict(torch.load(path)# 将其中的层直接替换为我们需要的层即可 model.fc = nn.Linear(2048,200)
resnext101_32x4d_gn-ac3bb84e.pth
查看来源:https://download.openmmlab.com/pretrain/third_party/resnext101_32x4d_gn-ac3bb84e.pthres2net101_v1d_26w_4s_mmdetv2-f0a600f9.pth
查看来源:https://download.openmmlab.com/pretrain/third_party/res2net101_v1d_26w_4s_mmdetv2-f0a600f9.pthuniversenet50_gfl_fp16_4x4_mstrain_480_960_2x_coco_20200729_epoch_24-c9308e66.pth
查看来源:https://github.com/shinya7y/UniverseNet/releases/download/20.07/universenet50_gfl_fp16_4x4_mstrain_480_960_2x_coco_20200729_epoch_24-c9308e66.pth