关于yolov5l_fusion_transformerx3_llvip.yaml网络结构层的问题

[h6h6h6king]

backbone:

[from, number, module, args]

Two Stream

[
######### TransformerBlock One #############
# stream one
[-1, 1, Focus, [64, 3]], # 0-P1/2
[-1, 1, Conv, [128, 3, 2]], # 1-P2/4
[-1, 3, C3, [128]], # 2-P2/4
[-1, 1, Conv, [256, 3, 2]], # 3-P3/8
[-1, 9, C3, [256]], # 4-P3/8
# stream two
[-4, 1, Focus, [64, 3]], # 5-P1/2
[-1, 1, Conv, [128, 3, 2]], # 6-P2/4
[-1, 3, C3, [128]], # 7-P2/4
[-1, 1, Conv, [256, 3, 2]], # 8-P3/8
[-1, 9, C3, [256]], # 9
请问这里第二流的输入为什么是-4，是因为在前向传播过程中使用的是-4作为条件所以无所谓吗

[eulerbaby123]

backbone:

[from, number, module, args]

Two Stream

[ ######### TransformerBlock One ############# # stream one [-1, 1, Focus, [64, 3]], # 0-P1/2 [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 [-1, 3, C3, [128]], # 2-P2/4 [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 [-1, 9, C3, [256]], # 4-P3/8 # stream two [-4, 1, Focus, [64, 3]], # 5-P1/2 [-1, 1, Conv, [128, 3, 2]], # 6-P2/4 [-1, 3, C3, [128]], # 7-P2/4 [-1, 1, Conv, [256, 3, 2]], # 8-P3/8 [-1, 9, C3, [256]], # 9 请问这里第二流的输入为什么是-4，是因为在前向传播过程中使用的是-4作为条件所以无所谓吗

你好，请问你解决了吗，我一直搞不懂这个是根据什么来判别是IR分支还是rgb分支，两个并行分支不应该都是接入的原始数据吗
，其他的yaml也是直接写的-4

[eulerbaby123]

backbone:

[from, number, module, args]

Two Stream

[ ######### TransformerBlock One ############# # stream one [-1, 1, Focus, [64, 3]], # 0-P1/2 [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 [-1, 3, C3, [128]], # 2-P2/4 [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 [-1, 9, C3, [256]], # 4-P3/8 # stream two [-4, 1, Focus, [64, 3]], # 5-P1/2 [-1, 1, Conv, [128, 3, 2]], # 6-P2/4 [-1, 3, C3, [128]], # 7-P2/4 [-1, 1, Conv, [256, 3, 2]], # 8-P3/8 [-1, 9, C3, [256]], # 9 请问这里第二流的输入为什么是-4，是因为在前向传播过程中使用的是-4作为条件所以无所谓吗

backbone:

[from, number, module, args]

Two Stream

[ ######### TransformerBlock One ############# # stream one [-1, 1, Focus, [64, 3]], # 0-P1/2 [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 [-1, 3, C3, [128]], # 2-P2/4 [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 [-1, 9, C3, [256]], # 4-P3/8 # stream two [-4, 1, Focus, [64, 3]], # 5-P1/2 [-1, 1, Conv, [128, 3, 2]], # 6-P2/4 [-1, 3, C3, [128]], # 7-P2/4 [-1, 1, Conv, [256, 3, 2]], # 8-P3/8 [-1, 9, C3, [256]], # 9 请问这里第二流的输入为什么是-4，是因为在前向传播过程中使用的是-4作为条件所以无所谓吗

你可以看看这个问题：The performance is different after replacing the RGB and ir stream #15，貌似好像就是讨论的这个东西

[getupearly11]

Hello, I also have a question about "-4". In the yolov5l_fusion_transformerx3_llvip.yaml file, "-4" means that the input comes from the output of the layer four steps back, that is, the output of [-1, 1, Conv, [128, 3, 2]], # 1-P2/4.

[eulerbaby123]

you can check it in the models/yolo_test.py, you can search -4 by used vscode or other idle,it has a sentence define such as if xx=-4, the income tensor from x[1] which means the ir modal, to understand what is x (rub omg+it img )you can check it in the data construction part, include if -4, in  any yaml it will use ir input， it not means the inputs from the formal 4 layers

…

---原始邮件--- 发件人: ***@***.***> 发送时间: 2025年4月11日(周五) 下午4:38 收件人: ***@***.***>; 抄送: "weijian ***@***.******@***.***>; 主题: Re: [DocF/multispectral-object-detection] 关于yolov5l_fusion_transformerx3_llvip.yaml网络结构层的问题 (Issue #84) Hello, I also have a question about "-4". In the yolov5l_fusion_transformerx3_llvip.yaml file, "-4" means that the input comes from the output of the layer four steps back, that is, the output of [-1, 1, Conv, [128, 3, 2]], # 1-P2/4. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***> getupearly11 left a comment (DocF/multispectral-object-detection#84) Hello, I also have a question about "-4". In the yolov5l_fusion_transformerx3_llvip.yaml file, "-4" means that the input comes from the output of the layer four steps back, that is, the output of [-1, 1, Conv, [128, 3, 2]], # 1-P2/4. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>