[proto] Added functional rotate_bounding_box op

[vfdev-5]

Related to #5514

Description:

• Added functional rotate_bounding_box op
• Added warning if expand=True and center is not None and disabled provided center
• Added tests

Results on synthetic images/bboxes:

Code

import numpy as np

import torch
import torchvision
from torchvision.prototype import features
from torchvision.prototype.transforms.functional import rotate_bounding_box, rotate_image_tensor


size = (64, 72)
# xyxy format
in_boxes = [
    [10, 15, 25, 35],
    [50, 5, 70, 22],
    [45, 46, 56, 62],
]

im1 = 255 * np.ones(size + (3, ), dtype=np.uint8)
for in_box in in_boxes:
    im1[in_box[1]:in_box[3], in_box[0]:in_box[2], :] = (127, 127, 127)
    
t_im1 = torch.tensor(im1).permute(2, 0, 1).view(1, 3, *size)

in_boxes = features.BoundingBox(
    in_boxes, format=features.BoundingBoxFormat.XYXY, image_size=size
)
    
angle = 22
center = None
expand = True

out_boxes = rotate_bounding_box(
    in_boxes, 
    in_boxes.format,
    in_boxes.image_size,
    angle,
    expand=expand,
    center=center
)
print(out_boxes)

t_im2 = rotate_image_tensor(t_im1, angle, expand=expand, center=center)


import cv2
import matplotlib.pyplot as plt
%matplotlib inline


plt.figure(figsize=(14, 10))

plt.subplot(1,2,1)
plt.title("Input image + bboxes")
r1 = t_im1[0, ...].permute(1, 2, 0).contiguous().cpu().numpy()
for in_box in in_boxes:    
    r1 = cv2.rectangle(r1, (in_box[0].item(), in_box[1].item()), (in_box[2].item(), in_box[3].item()), (255, 127, 0))
plt.imshow(r1)


plt.subplot(1,2,2)
plt.title("Output image + bboxes")
r2 = t_im2[0, ...].permute(1, 2, 0).contiguous().cpu().numpy()
for out_box in out_boxes:
    out_box = np.round(out_box.cpu().numpy()).astype("int32")
    r2 = cv2.rectangle(r2, (out_box[0], out_box[1]), (out_box[2], out_box[3]), (255, 127, 0), 0)
plt.imshow(r2)

Compare to D2:

TL;DR: matches the behaviour if expand=False and ~1 pixel difference otherwise.
This is due to cast to int : facebookresearch/detectron2@63ea900#diff-98378cbc4207da5ac829ceff03d62c48abbfc564911ac7d64c2f1d137b4f3474R130 and the way how affine matrix is computed : facebookresearch/detectron2@63ea900#diff-98378cbc4207da5ac829ceff03d62c48abbfc564911ac7d64c2f1d137b4f3474R171

Code

import numpy as np


size = image_size = (64, 64)
in_boxes = [
    [1, 1, 5, 5],
    [1, image_size[0] - 6, 5, image_size[0] - 2],
    [image_size[1] - 6, image_size[0] - 6, image_size[1] - 2, image_size[0] - 2],    
    [image_size[1] // 2 - 10, image_size[0] // 2 - 10, image_size[1] // 2 + 10, image_size[0] // 2 + 10],
]

im1 = 255 * np.ones(size + (3, ), dtype=np.uint8)
for in_box in in_boxes:
    im1[in_box[1]:in_box[3], in_box[0]:in_box[2], :] = (127, 127, 127)


angle = 34
expand = False
center = None if expand else [12, 23]

from detectron2.data.transforms import RotationTransform, AugmentationList
from detectron2.data.transforms import AugInput
import cv2

inpt = AugInput(im1, boxes=np.array(in_boxes, dtype="float32"))

augs = AugmentationList([RotationTransform(*size, angle, expand=expand, center=center, interp=cv2.INTER_NEAREST), ])
out = augs(inpt)
print(inpt.boxes)

import torch
import torchvision
from torchvision.prototype import features
from torchvision.prototype.transforms.functional import rotate_bounding_box, rotate_image_tensor


t_im1 = torch.tensor(im1).permute(2, 0, 1).view(1, 3, *size)

in_boxes = features.BoundingBox(
    in_boxes, format=features.BoundingBoxFormat.XYXY, image_size=size
)
    

out_boxes = rotate_bounding_box(
    in_boxes, 
    in_boxes.format,
    in_boxes.image_size,
    angle,
    expand=expand,
    center=center
)

from torchvision.prototype.transforms.functional import rotate_image_tensor

t_im2 = rotate_image_tensor(t_im1, angle, expand=expand, center=center)

import cv2
import matplotlib.pyplot as plt
%matplotlib inline


plt.figure(figsize=(14, 8))
plt.subplot(1,2,1)
plt.title("D2 input: image + bboxes")
r1 = im1
for in_box in in_boxes:    
    r1 = cv2.rectangle(r1, (in_box[0].item(), in_box[1].item()), (in_box[2].item(), in_box[3].item()), (255, 127, 0))
plt.imshow(r1)

ax = plt.subplot(1,2,2)
plt.title(f"D2 output: image + bboxes, angle={angle}, expand={expand}")
r2 = inpt.image
for out_box in inpt.boxes:
    out_box = np.round(out_box).astype("int32")
    r2 = cv2.rectangle(r2, (out_box[0], out_box[1]), (out_box[2], out_box[3]), (255, 127, 0), 1)
plt.imshow(r2)

plt.figure(figsize=(14, 8))

ax = plt.subplot(1,2,1)
plt.title("TV input: image + bboxes")
r1 = t_im1[0, ...].permute(1, 2, 0).contiguous().cpu().numpy()
for in_box in in_boxes:    
    r1 = cv2.rectangle(r1, (in_box[0].item(), in_box[1].item()), (in_box[2].item(), in_box[3].item()), (255, 127, 0))
plt.imshow(r1)


ax = plt.subplot(1,2,2)
plt.title(f"TV output: image + bboxes, angle={angle}, expand={expand}")
r2 = t_im2[0, ...].permute(1, 2, 0).contiguous().cpu().numpy()
for out_box in out_boxes:
    out_box = np.round(out_box.cpu().numpy()).astype("int32")
    r2 = cv2.rectangle(r2, (out_box[0], out_box[1]), (out_box[2], out_box[3]), (255, 127, 0), 0)
plt.imshow(r2)

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit d355c7b (more details on the Dr. CI page):

---

✅ None of the CI failures appear to be your fault 💚

---

• 3/3 broken upstream at merge base fdd3aab since Mar 21

---

🚧 3 ongoing upstream failures:

These were probably caused by upstream breakages that are not fixed yet.

• binary_linux_conda_py3.7_cu115 since Mar 21 (fbc8ea4)
  • 🔁 rerun
• binary_linux_conda_py3.10_cu115 since Mar 21 (fbc8ea4)
  • 🔁 rerun
• binary_linux_conda_py3.8_cu115 since Mar 21 (fbc8ea4)
  • 🔁 rerun

---

This comment was automatically generated by Dr. CI (expand for details).

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

[datumbox]

Thanks! LGTM with a couple of non-blocking nits for your consideration.