Skip to content
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

Some questions about the image features #21

Open
zefanZhang-cn opened this issue Oct 16, 2021 · 4 comments
Open

Some questions about the image features #21

zefanZhang-cn opened this issue Oct 16, 2021 · 4 comments

Comments

@zefanZhang-cn
Copy link

Hello, i want to ask about the image features which are extracted from the detectron. When i extract them, i find their shapes are (36,1024), but the features' shapes which are extracted by batra-mlp-lab are (36,2048). So my question is how to deal with this problem? Thanks.

@yuleiniu
Copy link
Owner

You could modify the option "model. img_feature_size" in the file "rva/configs/rva.yml" from 2048 to 1024.

@zefanZhang-cn
Copy link
Author

You could modify the option "model. img_feature_size" in the file "rva/configs/rva.yml" from 2048 to 1024.

感谢你的回复,但我还是想问下,为什么detectron提取出来的是1024,但模型使用的都是2048,这个是如何变化的呢?如果将文件中的 "model. img_feature_size" 改为1024会对结果有什么影响吗?因为我是在想利用以前的提取特征方法,将自己的数据集重新提取特征,跟原来batra-mlp-lab提取的特征保持一致,然后夹杂着一起使用,我看大家都是用的detectron来提取的特征,但我在使用时发现提取的特征是(36,1024)的,如果我直接利用全连接层将这个特征的shape改成(36,2048),然后放入到原来提取好的特征中,我怕这样会对我的结果产生影响。

@yuleiniu
Copy link
Owner

You could modify the option "model. img_feature_size" in the file "rva/configs/rva.yml" from 2048 to 1024.

感谢你的回复,但我还是想问下,为什么detectron提取出来的是1024,但模型使用的都是2048,这个是如何变化的呢?如果将文件中的 "model. img_feature_size" 改为1024会对结果有什么影响吗?因为我是在想利用以前的提取特征方法,将自己的数据集重新提取特征,跟原来batra-mlp-lab提取的特征保持一致,然后夹杂着一起使用,我看大家都是用的detectron来提取的特征,但我在使用时发现提取的特征是(36,1024)的,如果我直接利用全连接层将这个特征的shape改成(36,2048),然后放入到原来提取好的特征中,我怕这样会对我的结果产生影响。

请问你用的哪一个detectron的模型?根据batra-mlp-lab的代码https://github.com/batra-mlp-lab/visdial-challenge-starter-pytorch/blob/master/data/extract_features_detectron.py,提取特征的模型是基于faster_rcnn_x101的,你可以选取对应的模型提取自己的特征

@zefanZhang-cn
Copy link
Author

You could modify the option "model. img_feature_size" in the file "rva/configs/rva.yml" from 2048 to 1024.

感谢你的回复,但我还是想问下,为什么detectron提取出来的是1024,但模型使用的都是2048,这个是如何变化的呢?如果将文件中的 "model. img_feature_size" 改为1024会对结果有什么影响吗?因为我是在想利用以前的提取特征方法,将自己的数据集重新提取特征,跟原来batra-mlp-lab提取的特征保持一致,然后夹杂着一起使用,我看大家都是用的detectron来提取的特征,但我在使用时发现提取的特征是(36,1024)的,如果我直接利用全连接层将这个特征的shape改成(36,2048),然后放入到原来提取好的特征中,我怕这样会对我的结果产生影响。

请问你用的哪一个detectron的模型?根据batra-mlp-lab的代码https://github.com/batra-mlp-lab/visdial-challenge-starter-pytorch/blob/master/data/extract_features_detectron.py,提取特征的模型是基于faster_rcnn_x101的,你可以选取对应的模型提取自己的特征

是的,我就是选择的他所说的模型,但是特征shape不对应,不太理解。

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants