Torchvision bounding boxes do not match the images, becuase the bboxes are from the pre-cropped, pre-resized version.

🐛 Describe the bug

CelebA bounding boxes were calculated on the so called "in-the-wild" images, prior to cropping and resizing. But torchvision.datasets returns the version that is cropped to 178x218. So for example, on the ninth image, the bbox is outside the image size.

CODE TO REPRO

from torchvision import datasets

celeba = datasets.CelebA(root="./celeba", target_type="bbox", download=True, split="train")

print(celeba[8])

(<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=178x218>,
tensor([600, 274, 343, 475]))

Versions

collect_env.py crashed on me but here's the version:

Using Python 3.12.8 environment at: XXX
Name: torchvision
Version: 0.21.0
Location: XXX
Requires: numpy, pillow, torch
Required-by:

Start at the datasets.CelebA entry point and inspect how the stored bounding-box metadata relates to the cropped and resized image returned by celeba[8]. Reproduce the ninth-image example, then verify that the returned coordinates describe the object within the 178x218 image and add or update coverage for this case.