如何训练yolov_x_online的模型

[Eudian]

试了yolov_l_online, yolov_s_online的训练都跑不起来。

[YuHengsss]

These files are used for inferencing, please use config like yolov_l or yolov_s for training!

[Eudian]

So train with yolov_x.py, and inferencing with yolov_x_online.py, it goes in this way?

[YuHengsss]

yep!

…

---- 回复的原邮件 ---- 发件人 ***@***.***> 日期 2024年08月06日 14:02 收件人 ***@***.***> 抄送至 ***@***.***>***@***.***> 主题 Re: [YuHengsss/YOLOV] 如何训练yolov_x_online的模型 (Issue #92) So train with yolov_x.py, and inferencing with yolov_x_online.py, it goes in this way? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[Johnny-Lin]

class MSA_yolov_online(nn.Module):

def __init__(self, dim,out_dim, num_heads=4, qkv_bias=False, attn_drop=0.,scale=25):
    super().__init__()
    self.msa = Attention_msa_online(dim,num_heads,qkv_bias,attn_drop,scale=scale)
    self.linear1 = nn.Linear(2 * dim,2 * dim)
    self.linear2 =  nn.Linear(4 * dim,out_dim)

def ave_pooling_over_ref(self,features,sort_results):
    key_feature = features[0]
    support_feature = features[0]
    if not self.training:
        sort_results = sort_results.to(features.dtype)
    soft_sim_feature = (sort_results@support_feature)#.transpose(1, 2)#torch.sum(softmax_value * most_sim_feature, dim=1)
    cls_feature = torch.cat([soft_sim_feature,key_feature],dim=-1)

    return cls_feature

那这里推理用到的参数是怎么来的 ？