Skeleton for rotated bounding box tutorial

Author: scotts

docs/source/conf.py

@@ -87,6 +87,7 @@ def __init__(self, src_dir):
         "plot_transforms_illustrations.py",
         "plot_transforms_e2e.py",
         "plot_cutmix_mixup.py",
+        "plot_rotated_box_transforms.py",
         "plot_custom_transforms.py",
         "plot_tv_tensors.py",
         "plot_custom_tv_tensors.py",

gallery/assets/leaning_tower.jpg

@@ -0,0 +1,111 @@
+"""
+===============================================================
+Transforms on Rotated Bounding Boxes
+===============================================================
+
+Introduction.
+
+First, some code to set everything up.
+"""
+
+# %%
+from PIL import Image
+from pathlib import Path
+import matplotlib.pyplot as plt
+
+
+import torch
+import torchvision.transforms.functional as F
+from torchvision import tv_tensors
+from torchvision.transforms import v2
+from torchvision.utils import draw_bounding_boxes
+from helpers import plot
+
+plt.rcParams["figure.figsize"] = [10, 5]
+plt.rcParams["savefig.bbox"] = "tight"
+
+# if you change the seed, make sure that the randomly-applied transforms
+# properly show that the image can be both transformed and *not* transformed!
+torch.manual_seed(0)
+
+# If you're trying to run that on Colab, you can download the assets and the
+# helpers from https://github.com/pytorch/vision/tree/main/gallery/
+orig_img = Image.open(Path('../assets') / 'leaning_tower.jpg')
+
+# %%
+# Rotated Bounding Boxes
+# ----------------------
+# Brief intro into what rotated bounding boxes are. Brief description of the
+# image.
+
+orig_box = tv_tensors.BoundingBoxes(
+    [
+        [860.0, 1100, 570, 1840, -7],
+    ],
+    format="CXCYWHR",
+    canvas_size=(orig_img.size[1], orig_img.size[0]),
+    clamping_mode="hard",
+)
+# TODO: why is this necessary?
+orig_box = v2.ConvertBoundingBoxFormat("xyxyxyxy")(orig_box)
+
+plot([(orig_img, orig_box)])
+
+# %%
+# Image Rotation
+# ---------------
+# We can rotate the image itself, and the already rotated bounding boxes are
+# rotated appropriately.
+
+out_img, out_box = v2.RandomRotation(degrees=(0, 180), expand=True)(orig_img, orig_box)
+plot([(out_img, out_box)])
+
+# %%
+# Image Padding
+# -------------
+# The rotated bounding boxes also respect padding transforms.
+padded_imgs_and_boxes = [
+    v2.Pad(padding=padding)(orig_img, orig_box) for padding in (10, 30, 50, 100)
+]
+plot([(orig_img, orig_box)] + padded_imgs_and_boxes)
+
+# %%
+# Image Resizing
+# --------------
+# The rotated bounding boxes are resized along with the image.
+resized_imgs_and_boxes = [v2.Resize(size=size)(orig_img, orig_box) for size in (30, 50, 100, orig_img.size)]
+plot([(orig_img, orig_box)] + resized_imgs_and_boxes)
+
+# %%
+# Image Rotation
+# --------------
+rotater = v2.RandomRotation(degrees=(0, 180))
+rotated_imgs = [rotater((orig_img, orig_box)) for _ in range(4)]
+plot([(orig_img, orig_box)] + rotated_imgs)
+
+# %%
+# Elastic Transform
+# -----------------
+plot([v2.ElasticTransform(alpha=250.0)(orig_img, orig_box)])
+
+# %%
+# Clamping Modes
+# --------------
+# Explain hard and soft, with appropriate links to documentation. Talk about
+# defaults. Link to to-be-written-tutorial on mode-setting in general.
+soft_box = orig_box.clone()
+soft_box.clamping_mode = "soft"
+
+hard_center_crops_and_boxes = [
+    v2.CenterCrop(size=size)(orig_img, orig_box)
+    for size in (800, 1200, 2000, orig_img.size)
+]
+
+soft_center_crops_and_boxes = [
+    v2.CenterCrop(size=size)(orig_img, soft_box)
+    for size in (800, 1200, 2000, orig_img.size)
+]
+
+plot([[(orig_img, orig_box)] + hard_center_crops_and_boxes,
+      [(orig_img, soft_box)] + soft_center_crops_and_boxes])
+