Add bounding box converter #30852
Add bounding box converter #30852
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
There was a problem hiding this comment.
Thanks for working on this!
I haven't done a full review e.g. checking boring things like docstring formats, just looking at the overall structure.
This is really nicely structured and handled. I don't have any major comments on the overall logic - it's incredibly clean. Thanks for taking the time to write out comprehnsive tests too. Only major comment is to not have alias for the different formats, in particular for datasets like coco, it adds unnecessary complexity; I'm less sure about e.g. xyxy -> absolute_xyxy.
| @@ -0,0 +1,523 @@ | |||
| import logging | |||
There was a problem hiding this comment.
Missing copyright header
| return norm_annotation | ||
| """Convert annotation boxes from absolute xyxy (pascal voc) to relative xcycwh (yolo) format.""" | ||
| if "boxes" in annotation: | ||
| annotation = annotation.copy() # shallow copy |
There was a problem hiding this comment.
What's the purpose of the copy here? Only annotation["boxes"] is being passed to convert_boxes and is being set directly to annotation["boxes"] on the call
| ArrayType = Union["torch.Tensor", np.ndarray] | ||
|
|
||
|
|
||
| SUPPORTED_BOX_FORMATS = [ |
There was a problem hiding this comment.
I'd have this as an enum
| or most likely as a list of 2D arrays/tensors. | ||
|
|
||
| Supported input/output bounding box formats: | ||
| - `absolute_xyxy` (aliases: `pascal_voc`, `xyxy`): [x_min, y_min, x_max, y_max] |
There was a problem hiding this comment.
I'd rather we didn't try to handle different aliases and just keep it simple to e.g. xyxy
| input_format: str, | ||
| output_format: str, | ||
| image_size: Optional[Union[ArrayType, List, Tuple]] = None, | ||
| check: Optional[str] = "warn", |
There was a problem hiding this comment.
I can see "raise" never being used. I like the logic, but think we might be engineering something which we'll never need
| return _convert_boxes_arrays_2d(boxes, input_format, output_format, image_size, check) | ||
|
|
||
| # Recursive approach. | ||
| elif depth == 1 and isinstance(boxes, (list, tuple) or is_array_type(boxes)): |
There was a problem hiding this comment.
Regarding the isinstance(boxes, (list, tuple) or is_array_type(boxes)) check - what are the other types boxes could be here?
What does this PR do?
Intorduces
output_bbox_formatparameter forpostprocess_for_object_detection.Adds the
convert_boxesfunction to convert boxes from one format to another, e.g. from YOLO to Pascal VOC.np.array,torch.tensor,list,tuple(could be easily extended tojaxandtf)box=[x_min, y_min, x_max, y_max, color, class, ...]yolo,cocoSupported input/output bounding box formats:
-
absolute_xyxy(aliases:pascal_voc,xyxy): [x_min, y_min, x_max, y_max]-
absolute_xywh(aliases:coco,xywh): [x_min, y_min, width, height]-
absolute_xcycwh: [center_x, center_y, width, height]-
relative_xyxy(aliases:albumentations): [x_min, y_min, x_max, y_max] normalized to [0, 1] by image size-
relative_xywh: [x_min, y_min, width, height] normalized to [0, 1] by image size-
relative_xcycwh(aliases:yolo,xcycwh): [center_x, center_y, width, height] normalized to [0, 1] by image sizeBefore submitting
Pull Request section?
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
@amyeroberts could you please look at this draft, looks a bit overсomplicated, probably we should drop some functionality in favor to simplification.