New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
how to train model on vl_cmu_cd dataset? #13
Comments
This problem is about the input tensor shape of the ground truth. Please confirm your input gt's shape. |
The gt's shape should be (B 1 H W). |
thanks for reply,which format of vl_cmu_cd dataset should be selected ,the gt**.png or maks**.png? |
Please covert the gt.png in vl_cmu_cd dataset into mask.png which denotes the change map. |
|
It seems vlcmucd only has semantic labels instead of CD labels. You need to generate CD labels before training. Please review it's introduction in their paper. |
yes, i have convert the dataset to binary,and, the code is running.but, the new error is the following:Traceback (most recent call last): During handling of the above exception, another exception occurred: Traceback (most recent call last): During handling of the above exception, another exception occurred: Traceback (most recent call last): During handling of the above exception, another exception occurred: Traceback (most recent call last): |
Did u open visdom? |
Thank you for your guidance.The code is already running.i use test.txt as val.txt in vl_cmu_cd dataset.but the value of f1 is always 0. |
Something must be wrong. You may carefully check your input and gt.Please make sure your the final input gt value is between 0 and 1. |
Thank you for your reply. There are 11 classes in the vl_cmu_cd dataset, which means that the label values are the eleven values of [0,1]. Is that right? |
You are welcome. Hope u will contact me again if u have further questions. |
It seems this code are redundant. Please delete it. |
Thanks for ur reminding, I will update the code later. |
|
I'm so glad that my work helps you. Good luck. |
Hello, That's exciting news.I have developed my own model using open source data. It looks pretty good.However, I want to make my own data set, but I don't know how to set the label value of the data set. could you give me some guidance? |
The label values of semantic objects are all up to u. In a general way, for CD task, labels are set [0, 1] or [0, 255]. If u want to label ur own dataset like the setting in VL-CMU-CD, only give a specific value for each semantic category. The label tool is introduced to use 'labelme'. |
It's ok. |
Ok, thank you for your guidance. |
long time no see, how are you lately? I am planning to do some changedetection projects recently, I would like to ask you, is there a relationship between the accuracy of detection and the depth of the network? |
1 similar comment
long time no see, how are you lately? I am planning to do some changedetection projects recently, I would like to ask you, is there a relationship between the accuracy of detection and the depth of the network? |
Within limits,it is. |
ok, i got it. |
thanks for sharing your great work.when i use this code train my model like vl_cmu_cd format,but it did't work,the error is the following:
return _VF.broadcast_tensors(tensors) # type: ignore[attr-defined]
RuntimeError: The size of tensor a (512) must match the size of tensor b (3) at non-singleton dimension 4
look forward for your reply
The text was updated successfully, but these errors were encountered: