上篇文章介绍了Detection任务的预测返回结果内容，现在这篇文章介绍OBB(边界框检测)任务的预测返回结果。相比标准物体检测，OBB引入了一个额外的角度来更准确地定位图像中的物体，预测输出是一组精确包围图像中物体的旋转边界框。当物体以不同角度出现时，定向边界框尤其有用，例如在航空图像中，传统的轴对齐边界框可能会包含不必要的背景。
YOLO-obb 预测返回的是Python list类型的 Results 对象，包含的数据项很多，结构比较复杂，本文进行详细介绍。
Ultralytics官方文档的相关内容
根据Ultralytics官方文档的相关内容： YOLO-obb 推理返回的是Python list类型的 Results 对象。
from ultralytics import YOLO
# Load a model
model = YOLO("yolo11n-obb.pt") # load an official model
#model = YOLO("path/to/best.pt") # load a custom model
# Predict with the model
results = model("https://ultralytics.com/images/boats.jpg") # predict on an image
# Process results list
for result in results:
xywhr = result.obb.xywhr # center-x, center-y, width, height, angle (radians)
xyxyxyxy = result.obb.xyxyxyxy # polygon format with 4-points
names = [result.names[cls.item()] for cls in result.obb.cls.int()] # class name of each box
confs = result.obb.conf # confidence score of each box
obb = result.obb # Oriented boxes object for OBB outputs
下面结合实例进行具体分析。
Results实例
我们直接使用预训练模型yolo11n-obb.pt对boats.jpg图像进行检测。
检测完成后，输出的图像：
只输入一张图片，results列表只有一个值results[0]
print(results[0])：
boxes: None
keypoints: None
masks: None
names: {0: 'plane', 1: 'ship', 2: 'storage tank', 3: 'baseball diamond', 4: 'tennis court', 5: 'basketball court', 6: 'ground track field', 7: 'harbor',
8: 'bridge', 9: 'large vehicle', 10: 'small vehicle', 11: 'helicopter', 12: 'roundabout', 13: 'soccer ball field', 14: 'swimming pool'}
obb: ultralytics.engine.results.OBB object
orig_img: array([[[ 40, 68, 103],
[ 29, 56, 90],
[ 14, 36, 64],
...,
[ 86, 100, 94],
[ 88, 102, 96],
[ 92, 106, 100]],
...,
[[ 61, 59, 51],
[ 64, 62, 54],
[ 63, 61, 53],
...,
[ 24, 32, 9],
[ 24, 32, 9],
[ 24, 32, 9]]], dtype=uint8)
orig_shape: (1080, 1920)
path: 'D:\\work3\\YOLO-Parklot\\data\\images\\boats.jpg'
probs: None
save_dir: 'runs\\obb\\predict'
speed: {'preprocess': 175.00819999258965, 'inference': 2622.4026000127196, 'postprocess': 614.9457000137772}
print(result[0].obb)
cls: tensor([1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 7., 1., 7., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 7., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 7., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 7., 1., 1., 1., 1., 1., 1.])
conf: tensor([0.8425, 0.8336, 0.8319, 0.8297, 0.8263, 0.8244, 0.8243, 0.8218, 0.8197, 0.8181, 0.8177, 0.8162, 0.8147, 0.8134, 0.8132, 0.8122, 0.8106, 0.8105, 0.8101, 0.8097, 0.8094, 0.8076, 0.8074, 0.8067, 0.8067, 0.8067, 0.8065, 0.8061, 0.8059, 0.8055, 0.8052, 0.8035, 0.8026, 0.8014, 0.7996, 0.7991, 0.7979, 0.7965, 0.7965, 0.7909, 0.7904, 0.7895, 0.7891, 0.7887, 0.7887, 0.7884, 0.7861, 0.7851, 0.7846, 0.7845, 0.7843, 0.7841, 0.7840, 0.7838, 0.7835, 0.7817, 0.7807, 0.7806, 0.7804, 0.7777, 0.7772, 0.7754, 0.7744, 0.7742, 0.7736, 0.7720, 0.7711, 0.7696, 0.7692, 0.7669, 0.7664, 0.7652, 0.7638, 0.7634, 0.7625, 0.7595, 0.7591, 0.7493, 0.7489, 0.7480, 0.7475, 0.7473, 0.7471, 0.7468, 0.7462, 0.7456, 0.7430, 0.7369, 0.7368, 0.7358, 0.7353, 0.7345, 0.7341, 0.7340, 0.7278, 0.7273, 0.7270, 0.7266, 0.7253, 0.7252, 0.7224, 0.7213, 0.7201, 0.7196, 0.7160, 0.7131, 0.7102, 0.7079, 0.7063, 0.7025, 0.7014, 0.7007, 0.6997, 0.6957, 0.6889, 0.6873, 0.6849, 0.6839, 0.6818, 0.6815, 0.6803, 0.6785, 0.6743, 0.6734, 0.6732, 0.6727, 0.6726, 0.6705, 0.6690, 0.6682, 0.6673, 0.6632, 0.6568, 0.6557, 0.6507, 0.6438, 0.6435, 0.6367, 0.6336, 0.6320, 0.6290, 0.6271, 0.6255, 0.6169, 0.6096, 0.5993, 0.5991, 0.5949, 0.5922, 0.5855, 0.5845, 0.5666, 0.5582, 0.5355, 0.5275, 0.5091, 0.4968, 0.4921, 0.4763, 0.4447, 0.4364, 0.4333, 0.4177, 0.4068, 0.4049, 0.3979, 0.3959, 0.3898, 0.3788, 0.3784, 0.3600, 0.3593, 0.3432, 0.3253, 0.2982, 0.2772, 0.2572])
data: tensor([[3.2304e+02, 4.4219e+02, 7.6033e+01, ..., 3.7826e-01, 8.4254e-01, 1.0000e+00],
[1.7504e+03, 6.9368e+02, 1.0658e+02, ..., 4.1058e-01, 8.3363e-01, 1.0000e+00],
[1.8187e+03, 5.4377e+02, 9.5871e+01, ..., 4.4453e-01, 8.3185e-01, 1.0000e+00],
...,
[4.8026e+02, 1.0163e+03, 6.8284e+01, ..., 1.3577e+00, 2.9824e-01, 1.0000e+00],
[1.7286e+03, 2.3165e+01, 5.7848e+01, ..., 3.3524e-01, 2.7723e-01, 1.0000e+00],
[4.4217e+02, 3.1912e+01, 4.7885e+01, ..., 3.7069e-01, 2.5719e-01, 1.0000e+00]])
id: None
is_track: False
orig_shape: (1080, 1920)
shape: torch.Size([177, 7])
xywhr: tensor([[3.2304e+02, 4.4219e+02, 7.6033e+01, 2.7323e+01, 3.7826e-01],
[1.7504e+03, 6.9368e+02, 1.0658e+02, 3.2217e+01, 4.1058e-01],
[1.8187e+03, 5.4377e+02, 9.5871e+01, 2.9535e+01, 4.4453e-01],
[1.6765e+03, 8.6366e+02, 1.1517e+02, 3.3924e+01, 4.2344e-01],
...,
[4.2051e+02, 1.0421e+03, 8.3692e+01, 2.5618e+01, 1.2571e+00],
[4.8026e+02, 1.0163e+03, 6.8284e+01, 2.2957e+01, 1.3577e+00],
[1.7286e+03, 2.3165e+01, 5.7848e+01, 2.0636e+01, 3.3524e-01],
[4.4217e+02, 3.1912e+01, 4.7885e+01, 1.6795e+01, 3.7069e-01]])
xyxy: tensor([[ 2.8267e+02, 4.1545e+02, 3.6342e+02, 4.6892e+02],
[ 1.6951e+03, 6.5764e+02, 1.8057e+03, 7.2972e+02],
[ 1.7691e+03, 5.0982e+02, 1.8684e+03, 5.7771e+02],
[ 1.6170e+03, 8.2453e+02, 1.7359e+03, 9.0278e+02],
...,
[ 3.9541e+02, 9.9833e+02, 4.4561e+02, 1.0858e+03],
[ 4.6182e+02, 9.8054e+02, 4.9870e+02, 1.0521e+03],
[ 1.6979e+03, 3.9058e+00, 1.7593e+03, 4.2425e+01],
[ 4.1681e+02, 1.5412e+01, 4.6752e+02, 4.8413e+01]])
xyxyxyxy: tensor([[[ 353.3254, 468.9211],
[ 363.4159, 443.5294],
[ 292.7574, 415.4503],
[ 282.6669, 440.8420]],
[[1792.7980, 729.7223],
[1805.6569, 700.1832],
[1707.9366, 657.6438],
[1695.0778, 687.1829]],
[[1855.6692, 577.7137],
[1868.3701, 551.0494],
[1781.8164, 509.8216],
[1769.1155, 536.4860]],
...,
[[ 476.2670, 1052.1387],
[ 498.7041, 1047.2825],
[ 484.2596, 980.5439],
[ 461.8225, 985.4000]],
[[1752.4958, 42.4249],
[1759.2852, 22.9373],
[1704.6580, 3.9058],
[1697.8687, 23.3935]],
[[ 461.4397, 48.4127],
[ 467.5239, 32.7583],
[ 422.8915, 15.4118],
[ 416.8073, 31.0662]]])
xyxyxyxyn: tensor([[[0.1840, 0.4342],
[0.1893, 0.4107],
[0.1525, 0.3847],
[0.1472, 0.4082]],
[[0.9337, 0.6757],
[0.9404, 0.6483],
[0.8896, 0.6089],
[0.8829, 0.6363]],
[[0.9665, 0.5349],
[0.9731, 0.5102],
[0.9280, 0.4721],
[0.9214, 0.4967]],
...,
[[0.2481, 0.9742],
[0.2597, 0.9697],
[0.2522, 0.9079],
[0.2405, 0.9124]],
[[0.9128, 0.0393],
[0.9163, 0.0212],
[0.8878, 0.0036],
[0.8843, 0.0217]],
[[0.2403, 0.0448],
[0.2435, 0.0303],
[0.2203, 0.0143],
[0.2171, 0.0288]]])
Results数据解读
Result对象整体结构
使用model.predict()对一批图像进行预测时，会返回一个列表，列表中的每个元素都是一个Result对象，每个Result对象代表了对一张输入图像的预测结果。
Result对象的主要属性
原始图像属性
orig_img：原始的输入图像，输入图像的原始像素信息，数据类型为uint8（dtype=uint8）- ‘orig_shape’:，图像的原始高度和宽度。
- ‘path’:图像的原始路径
OBB属性
obb：OBB对象属性，包含了图像中检测到的所有边界框的信息，包括边界框的坐标、置信度以及类别索引等。
obb具有以下属性：
cls：检测到边界框的分类index。每个index对应的名称在’names’: 属性中（本例中直接使用预训练模型，names在dota.yaml中定义）。
conf: 每个边界框的置信度
xywhr: 边界框在原图中的中心坐标值和宽高及角度
xyxy: 边界框在原图中的左上和右下两个点的坐标
xyxyxyxy：边界框的四个角点坐标
xyxyxyxyn: xyxyxyxy的归一化值
其他属性
本例中只有obb属性的数据，没有’boxes’, ‘masks’, ‘probs’, ‘keypoints’, 。
speed：前置处理、推理、后置处理各个阶段的处理时间
names: 索引对应的分类名称
path：图像的原始路径
save_dir: 结果数据保存路径
结果数据调用
可以使用以下方式展示和保存处理后的图像：
from PIL import Image
from ultralytics import YOLO
# Load a model
model = YOLO("yolo11n-obb.pt") # load an official model
# Predict with the model
results = model("https://ultralytics.com/images/boats.jpg") # predict on an image
for i, r in enumerate(results):
# Plot results image
im_bgr = r.plot() # BGR-order numpy array
im_rgb = Image.fromarray(im_bgr[..., ::-1]) # RGB-order PIL image
# Show results to screen (in supported environments)
r.show()
# Save results to disk
r.save(filename=f"results{i}.jpg")
扫一扫
984
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
