[Feature] Create specific-sized box

[alicera]

I want to create a box with (256x256).
How can I do with labelme?

[wkentaro]

I'm not sure if there are many needs for this.

[alicera]

I have a idea.
Assume the label is 1 class person.
I label many images and I find some images have the ambiguous status.
The ambiguous status define fully black person, half person, and occlude etc.
So I want to create a specific-sized box to cut it. And it can avoid the ambiguous status.

It is inspired
ultralytics/yolov5#1423

[wkentaro]

What do you mean by “cut it”?

[alicera]

REF ：https://github.com/biubug6/Pytorch_Retinaface

The blue circle is a ambiguous status. If I label it, it maybe affect the detect result.

So I can crop it. And label it

[wkentaro]

So you mean you want to "ignore it" by "cut it"?
Why not just use --labelflags?
https://github.com/wkentaro/labelme/tree/main/examples/instance_segmentation#annotation

labelme data_annotated --labels labels.txt --nodata --labelflags '{.*: [occluded, truncated], person: [male]}'

[alicera]

I dont want to write the info [occluded] into the json.And it will contain the original image not the crop. It is not the suitable for me.
The crop have another advantage.For example, my data is 4:3(w:h), but I search some 16:9 data. I can crop it with 4:3. Let the data contain 4:3 ratio.