When I try to calculate the anchor boxes, I get an error due to missing the full size of the image. - #7
Conversation
… because ImageLab doesn't add the total image size to the XML. Now, the program checks if the figure size exists; if not, it iterates through all the figures and gets their size, which is necessary to perform the calculation. For large datasets, the process is slow, but it works perfectly.
|
If your dataset does not have the width/height metadata, you can add it with
This was added precisely for this use case in |
|
So, to be clear, the reason why this information should be in the dataset is because, generally, you want to try several settings to find the best bounding boxes. Having to load all the images from the dataset everytime makes it very slow to compute the boxes. If that solves your issue, you can now close the PR. |
|
This makes me think that we should probably use these in What do you think @davisking, would like a PR that does that? |
|
The biggest difficulty was that when you don’t have the full size of the figure, the program crashes without indicating the reason. |
|
Ah, I should probably improve the error message, then I will do that at some point later, let me add an issue so that I don't forget (or people can publicly shame me for not doing it) |
|
@arrufat yeah if you want to make a PR that makes |
When I try to calculate the boxes, an error occurs. This error occurs because ImageLab doesn't add the total image size to the XML. Now, the program checks if the figure size exists; if not, it iterates through all the figures and gets their size, which is necessary to perform the calculation. For large datasets, the process is slow, but it works perfectly.