Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -795,6 +795,15 @@ Each instance has scalar attributes (canvas) and sequence attributes (elements).
|
|
795 |
<PIL.PngImagePlugin.PngImageFile image mode=RGBA size=256x256>]}
|
796 |
```
|
797 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
798 |
### Data Fields
|
799 |
|
800 |
In the following, categorical fields are shown as `categorical` type, but the actual storage is `int64`.
|
@@ -851,6 +860,50 @@ The Crello dataset has 3 splits: train, validation, and test. The current split
|
|
851 |
| test | 2371 |
|
852 |
|
853 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
854 |
## Dataset Creation
|
855 |
|
856 |
### Curation Rationale
|
|
|
795 |
<PIL.PngImagePlugin.PngImageFile image mode=RGBA size=256x256>]}
|
796 |
```
|
797 |
|
798 |
+
To get a label for categorical values, use the `int2str` method:
|
799 |
+
|
800 |
+
```python
|
801 |
+
key = "font"
|
802 |
+
example = dataset[0]
|
803 |
+
|
804 |
+
dataset.features[key].int2str(example[key])
|
805 |
+
```
|
806 |
+
|
807 |
### Data Fields
|
808 |
|
809 |
In the following, categorical fields are shown as `categorical` type, but the actual storage is `int64`.
|
|
|
860 |
| test | 2371 |
|
861 |
|
862 |
|
863 |
+
### Visualization
|
864 |
+
|
865 |
+
Each example can be visualized in the following approach. Note the following does not guarantee a similar appearance to the original template.
|
866 |
+
|
867 |
+
```python
|
868 |
+
def render(dataset: datasets.Dataset, example: Dict[str, Any], max_size: float=512.) -> bytes:
|
869 |
+
"""Render parsed sequence example onto an image and return as PNG bytes."""
|
870 |
+
canvas_width = int(dataset.features["canvas_width"].int2str(example["canvas_width"]))
|
871 |
+
canvas_height = int(dataset.features["canvas_height"].int2str(example["canvas_height"]))
|
872 |
+
|
873 |
+
scale = min(1.0, max_size / canvas_width, max_size / canvas_height)
|
874 |
+
|
875 |
+
surface = skia.Surface(int(scale * canvas_width), int(scale * canvas_height))
|
876 |
+
with surface as canvas:
|
877 |
+
canvas.scale(scale, scale)
|
878 |
+
for index in range(example["length"]):
|
879 |
+
pil_image = example["image"][index]
|
880 |
+
image = skia.Image.frombytes(
|
881 |
+
pil_image.convert('RGBA').tobytes(),
|
882 |
+
pil_image.size,
|
883 |
+
skia.kRGBA_8888_ColorType)
|
884 |
+
left = example["left"][index] * canvas_width
|
885 |
+
top = example["top"][index] * canvas_height
|
886 |
+
width = example["width"][index] * canvas_width
|
887 |
+
height = example["height"][index] * canvas_height
|
888 |
+
rect = skia.Rect.MakeXYWH(left, top, width, height)
|
889 |
+
|
890 |
+
angle = example["angle"][index]
|
891 |
+
if angle != 0:
|
892 |
+
degree = 180. * angle / np.pi
|
893 |
+
canvas.save()
|
894 |
+
canvas.rotate(degree, left + width / 2., top + height / 2.)
|
895 |
+
|
896 |
+
canvas.drawImageRect(image, rect)
|
897 |
+
if angle != 0:
|
898 |
+
canvas.restore()
|
899 |
+
|
900 |
+
image = surface.makeImageSnapshot()
|
901 |
+
with io.BytesIO() as f:
|
902 |
+
image.save(f, skia.kPNG)
|
903 |
+
return f.getvalue()
|
904 |
+
```
|
905 |
+
|
906 |
+
|
907 |
## Dataset Creation
|
908 |
|
909 |
### Curation Rationale
|