PatchViewer

class ibb.widgets.PatchViewer(**kwargs: Any)

This widget works in a similar way as the BramboxViewer, but is meant to be used with large images and splits it in smaller patches with overlap.
It’s arguments work a lot like brambox’s brambox.util.BoxDrawer class.

Parameters

• images (callable or dict-like object) – A way to get the image or path to the image from the image labels in the dataframe

• boxes (pandas.DataFrame) – Bounding boxes to draw

• patch (int or tuple of int) – Width and height of the patch

• overlap (int, or tuple of int, optional) – width and height overlap between patches; Default computed automatically

• label (pandas.Series) – Label to write above the boxes; Default class_label (confidence)

• color (pandas.Series) – Color to use for drawing; Default every class_label will get its own color, up to 10 labels

• size (pandas.Series) – Thickness of the border of the bounding boxes; Default 3

• alpha (pandas.Series) – Alpha fill value of the bounding boxes; Default 00

• pad (int, float or tuple of 2 int/float) – Padding to add around the width and height (see Note); Default 10

• **kwargs (dict) – Extra keyword arguments that will be passed to Viewer

Note

If the images argument is callable, the image or path to the image will be retrieved in the following way: >>> image = images(image_label)

Otherwise the image or path is retrieved as: >>> image = images[image_label]

Note

The overlap property can be one of 3 possible types: - int : A fixed pixel overlap is added to each side. - (int, int) : Separate pixel overlap for the width and height. - None : Automatically compute overlap so all patches contain full image data

Note

The label, color, size and alpha arguments can also be tacked on to the boxes dataframe as columns. They can also be a single value, which will then be used for each bounding box.
Basically, as long as you can assign the value as a new column to the dataframe, it will work.

Examples

Warning

Note that the widgets are not responding as we do not have a python backend running!

import brambox as bb
import ibb

# Load annotations
anno = bb.io.load('pandas', 'annotations.h5')
anno

	image	class_label	id	x_top_left	y_top_left	width	height	occluded	truncated	lost	difficult	ignore
0	000001	dog	NaN	48.0	240.0	147.0	131.0	0.0	1.0	False	False	False
1	000001	person	NaN	8.0	12.0	344.0	486.0	0.0	1.0	False	False	False
2	000002	train	NaN	139.0	200.0	68.0	101.0	0.0	0.0	False	False	False
3	000003	sofa	NaN	123.0	155.0	92.0	40.0	0.0	0.0	False	False	False
4	000003	chair	NaN	239.0	156.0	68.0	49.0	0.0	0.0	False	False	False
5	000004	car	NaN	13.0	311.0	71.0	51.0	0.0	0.0	False	False	False
6	000004	car	NaN	362.0	330.0	138.0	59.0	0.0	1.0	False	False	False
7	000004	car	NaN	235.0	328.0	99.0	47.0	0.0	0.0	False	False	False
8	000004	car	NaN	175.0	327.0	77.0	37.0	0.0	0.0	False	False	False
9	000004	car	NaN	139.0	320.0	50.0	39.0	0.0	0.0	False	False	False
10	000004	car	NaN	108.0	325.0	42.0	28.0	0.0	0.0	False	False	False
11	000004	car	NaN	84.0	323.0	37.0	27.0	0.0	0.0	False	False	False

v = ibb.PatchViewer(
    # Dictionary or Function to map image column to image paths
    lambda name: f'images/{name}.jpg',
    anno,
    patch=100,
    overlap=10,
)
v

# Load Detections
det = bb.io.load('pandas', 'detections.h5')
det

	image	class_label	id	x_top_left	y_top_left	width	height	confidence
0	000001	person	NaN	270.069856	55.980325	35.082728	99.642827	0.002556
1	000001	tvmonitor	NaN	5.429369	80.088432	26.879914	99.102836	0.001120
2	000001	person	NaN	273.802024	96.162503	32.399382	73.652089	0.034645
3	000001	person	NaN	318.425435	77.141560	28.912620	123.342065	0.008122
4	000001	person	NaN	352.354380	93.409190	9.711184	83.560586	0.001517
...	...	...	...	...	...	...	...	...
164	000004	car	NaN	67.649786	306.099818	296.283392	96.845572	0.001021
165	000004	motorbike	NaN	-35.525570	334.530390	342.759683	109.085184	0.001464
166	000004	car	NaN	6.774407	340.022931	330.145946	97.088520	0.001446
167	000004	person	NaN	41.884642	136.205545	339.365922	438.765929	0.001059
168	000004	bus	NaN	123.889300	143.956826	334.710195	428.265535	0.001212

169 rows × 8 columns

ibb.PatchViewer(
    # Dictionary or Function to map image column to image paths
    lambda name: f'images/{name}.jpg',
    det,

    # Set overlap to 100 and let overlap be computed automatically
    patch=100,

    # Scale border size, based on confidence levels
    size=det.confidence * 5,
)