Added spatial bbox helper (#171)

Summary:
- [x] I have read CONTRIBUTING.md to understand how to contribute to this repository :)

Computes the bbox that encloses a white box in a black background for any augmentation.

### Image
```bash
python -m unittest discover -s augly/tests/image_tests/ -p "*_test.py"
# Or `python -m unittest discover -s augly/tests/image_tests/ -p "*.py"` to run pytorch test too (must install `torchvision` to run)
```
```
Ran 82 tests in 53.014s

OK (skipped=5)
```

## Other testing

Colab notebook testing the bbox helper → https://colab.research.google.com/drive/1g_0I6f_bv4Wsna6l9jjZrOJ62a4dpu8U#scrollTo=yUczCe6FU9Bs

Pull Request resolved: #171

Reviewed By: jbitton

Differential Revision: D32951437

Pulled By: zpapakipos

fbshipit-source-id: 7b3532a5e679e08f60949dd29e8929806e1557e3

augly/image/utils/bboxes.py

@@ -2,10 +2,11 @@
 # Copyright (c) Facebook, Inc. and its affiliates.
 
 import math
-from typing import List, Optional, Tuple
+from typing import Callable, List, Optional, Tuple
 
 import augly.image.utils as imutils
 import numpy as np
+from PIL import Image, ImageDraw
 
 
 def crop_bboxes_helper(
@@ -444,6 +445,37 @@ def get_enclosing_bbox(
     )
 
 
+def spatial_bbox_helper(
+    bbox: Tuple[float, float, float, float],
+    src_w: int,
+    src_h: int,
+    aug_function: Callable,
+    **kwargs,
+) -> Tuple:
+    """
+    Computes the bbox that encloses the transformed bbox in the image transformed by
+    `aug_function`. This helper can be used to compute the transformed bbox for any
+    augmentation which doesn't affect the color of the source image (e.g. any spatial
+    augmentation).
+    """
+    dummy_image = Image.new("RGB", (src_w, src_h))
+    draw = ImageDraw.Draw(dummy_image)
+    draw.rectangle(
+        (bbox[0] * src_w, bbox[1] * src_h, bbox[2] * src_w, bbox[3] * src_h),
+        fill="white",
+    )
+
+    aug_image = aug_function(dummy_image, **kwargs)
+    aug_w, aug_h = aug_image.size
+    array_image = np.array(aug_image)
+
+    white_y, white_x, _ = np.where(array_image > 0)
+    min_x, max_x = np.min(white_x), np.max(white_x)
+    min_y, max_y = np.min(white_y), np.max(white_y)
+
+    return (min_x / aug_w, min_y / aug_h, max_x / aug_w, max_y / aug_h)
+
+
 def vflip_bboxes_helper(bbox: Tuple, **kwargs) -> Tuple:
     """
     Analogous to hflip, when the src image is vertically flipped, the bounding box also