咨询
gengyanlei opened this issue · comments
作者，你好，请问你这里有s2anet的ucas-aod的模型么
DAL模型 demo.py
Unexpected key(s) in state_dict: "cls_attention.channel_att.gate_c.gate_c_fc_0.weight", "cls_attention.channel_att.gate_c.gate_c_fc_0.bias", "cls_attention.channel_att.gate_c.gate_c_bn_1.weight", "cls_attention.channel_att.gate_c.gate_c_bn_1.bias", "cls_attention.channel_att.gate_c.gate_c_bn_1.running_mean", "cls_attention.channel_att.gate_c.gate_c_bn_1.running_var", "cls_attention.channel_att.gate_c.gate_c_bn_1.num_batches_tracked", "cls_attention.channel_att.gate_c.gate_c_fc_final.weight", "cls_attention.channel_att.gate_c.gate_c_fc_final.bias", "cls_attention.spatial_att.branch1x1.conv.weight", "cls_attention.spatial_att.branch3x1.conv.weight", "cls_attention.spatial_att.branch1x3.conv.weight", "cls_attention.spatial_att.branch3x3.conv.weight", "cls_attention.spatial_att.conv.0.conv.weight", "cls_attention.spatial_att.conv.1.weight", "cls_attention.spatial_att.conv.1.bias", "cls_attention.spatial_att.conv.1.running_mean", "cls_attention.spatial_att.conv.1.running_var", "cls_attention.spatial_att.conv.1.num_batches_tracked", "cls_attention.spatial_att.conv.3.conv.weight", "reg_attention.channel_att.gate_c.gate_c_fc_0.weight", "reg_attention.channel_att.gate_c.gate_c_fc_0.bias", "reg_attention.channel_att.gate_c.gate_c_bn_1.weight", "reg_attention.channel_att.gate_c.gate_c_bn_1.bias", "reg_attention.channel_att.gate_c.gate_c_bn_1.running_mean", "reg_attention.channel_att.gate_c.gate_c_bn_1.running_var", "reg_attention.channel_att.gate_c.gate_c_bn_1.num_batches_tracked", "reg_attention.channel_att.gate_c.gate_c_fc_final.weight", "reg_attention.channel_att.gate_c.gate_c_fc_final.bias", "reg_attention.spatial_att.branch1x1.conv.weight", "reg_attention.spatial_att.branch3x1.conv.weight", "reg_attention.spatial_att.branch1x3.conv.weight", "reg_attention.spatial_att.branch3x3.conv.weight", "reg_attention.spatial_att.conv.0.conv.weight", "reg_attention.spatial_att.conv.1.weight", "reg_attention.spatial_att.conv.1.bias", "reg_attention.spatial_att.conv.1.running_mean", "reg_attention.spatial_att.conv.1.running_var", "reg_attention.spatial_att.conv.1.num_batches_tracked", "reg_attention.spatial_att.conv.3.conv.weight".
这样操作，才不会报错，但是检测效果不是很好。
请指教一下，tks
可能是权重加载错了，带attention的是CFC-Net的weight，DAL是不带的
你好，你给的权重里面就包含这个，不是我加载错了
抱歉现在才看到你的回复,确认了一下确实是放错权重了.这个是cfcnet的权重.
你可以直接load的时候跳过这些层就行,或者直接重新训练一下模型,readme写的比较详细应该也不难.
有问题可以邮箱联系,会回复得及时一点.
tks