The optflow results of Pretrained weights is all zeros

[llltttppp]

预训练的SegFlow.caffemodel模型中的score_res3_con_flow层权值全为０导致光流输出结果全是０，希望能提供正确的权值下载

In pretrained weights SegFlow.caffemodel, weights of layer("score_res3_con_flow") is all zeros, please upload a correct pretrained weights, thank you!

[fish8sank]

I have the same questions as llltttppp said,even when i used the pretrained flownet model to train my own data,the flow result turns out to be all zeros.
Need some help.

[lcf000000]

I have the same questions as llltttppp said,even when i used the pretrained flownet model to train my own data,the flow result turns out to be all zeros.
Need some help.

I use the pre-trained model which they provide to us, but the result is wrong, it can't segment anything.
Should I train the model on myself ?

[fish8sank]

I have the same questions as llltttppp said,even when i used the pretrained flownet model to train my own data,the flow result turns out to be all zeros.
Need some help.

I use the pre-trained model which they provide to us, but the result is wrong, it can't segment anything.
Should I train the model on myself ?

I think you should train the network with your own data

[lcf000000]

I have the same questions as llltttppp said,even when i used the pretrained flownet model to train my own data,the flow result turns out to be all zeros.
Need some help.

I use the pre-trained model which they provide to us, but the result is wrong, it can't segment anything.
Should I train the model on myself ?

I think you should train the network with your own data

I just run the demo.
$ python infer_DAVIS.py dog
So they just provide a pre-trained model instead of a well trained model?

[fish8sank]

I have the same questions as llltttppp said,even when i used the pretrained flownet model to train my own data,the flow result turns out to be all zeros.
Need some help.

I use the pre-trained model which they provide to us, but the result is wrong, it can't segment anything.
Should I train the model on myself ?

I think you should train the network with your own data

I just run the demo.
$ python infer_DAVIS.py dog
So they just provide a pre-trained model instead of a well trained model?

You mean you test the data they provide by using the model they gave?If so, there should be segmentation results.You can check infer_DAVIS.py, I assume there are something wrong in your demo code.

[lcf000000]

I have the same questions as llltttppp said,even when i used the pretrained flownet model to train my own data,the flow result turns out to be all zeros.
Need some help.

I use the pre-trained model which they provide to us, but the result is wrong, it can't segment anything.
Should I train the model on myself ?

I think you should train the network with your own data

I just run the demo.
$ python infer_DAVIS.py dog
So they just provide a pre-trained model instead of a well trained model?

You mean you test the data they provide by using the model they gave?If so, there should be segmentation results.You can check infer_DAVIS.py, I assume there are something wrong in your demo code.

Oh， I modifed deploy.prototxt file.
Now , it works well, thx very much !

[xieguotian]

I have the same questions as llltttppp said,even when i used the pretrained flownet model to train my own data,the flow result turns out to be all zeros.
Need some help.

I use the pre-trained model which they provide to us, but the result is wrong, it can't segment anything.
Should I train the model on myself ?

I think you should train the network with your own data

I just run the demo.
$ python infer_DAVIS.py dog
So they just provide a pre-trained model instead of a well trained model?

You mean you test the data they provide by using the model they gave?If so, there should be segmentation results.You can check infer_DAVIS.py, I assume there are something wrong in your demo code.

Oh， I modifed deploy.prototxt file.
Now , it works well, thx very much !

How to modify deploy.prototxt file ?

[lcf000000]

I have the same questions as llltttppp said,even when i used the pretrained flownet model to train my own data,the flow result turns out to be all zeros.
Need s

预训练的SegFlow.caffemodel模型中的score_res3_con_flow层权值全为０导致光流输出结果全是０，希望能提供正确的权值下载

In pretrained weights SegFlow.caffemodel, weights of layer("score_res3_con_flow") is all zeros, please upload a correct pretrained weights, thank you!

Did you solve this problem ? The flow branch always output zero, so it doesn't contribute to segmentation result actually.

[mingqizhang]

在训练的时候有个问题，分割分支和光流分支的训练数据是不一样的。但两个分支在后部分有特征层相互使用，所以在训练分割分支时输入是分割数据，光流分支参数固定，但后面给分割分支的光流特征还是要有的，所以光流分支的输入是什么呢，难道是同样的分割数据吗。同理在训练光流分支的时候，输入的光流数据，那分割之分的输入是什么呢？请问有人知道吗？

[ShristiDasBiswas]

I have the same questions as llltttppp said,even when i used the pretrained flownet model to train my own data,the flow result turns out to be all zeros.
Need s

预训练的SegFlow.caffemodel模型中的score_res3_con_flow层权值全为０导致光流输出结果全是０，希望能提供正确的权值下载
In pretrained weights SegFlow.caffemodel, weights of layer("score_res3_con_flow") is all zeros, please upload a correct pretrained weights, thank you!

Did you solve this problem ? The flow branch always output zero, so it doesn't contribute to segmentation result actually.

were you able to get it working? please share what changes you made.