Skip to content

Commit eb4afdd

Browse files
rlaalsrl0922jungnerdstevhliu
authored
[i18n-KO] Translated keypoint_detection.md to Korean (#36649)
* fix: manual edits * fix: manual edits * fix: manual edits * Update docs/source/ko/tasks/keypoint_detection.md Anchor lower modify Co-authored-by: Woojun Jung <[email protected]> * Update docs/source/ko/tasks/keypoint_detection.md connect letter Co-authored-by: Woojun Jung <[email protected]> * Update docs/source/ko/tasks/keypoint_detection.md modify to usual words Co-authored-by: Woojun Jung <[email protected]> * Update docs/source/ko/tasks/keypoint_detection.md modify extension word Co-authored-by: Steven Liu <[email protected]> * Update docs/source/ko/tasks/keypoint_detection.md modify to usual words Co-authored-by: Woojun Jung <[email protected]> * Update docs/source/ko/tasks/keypoint_detection.md modify to usual words Co-authored-by: Woojun Jung <[email protected]> * Update docs/source/ko/tasks/keypoint_detection.md modify to usual representation Co-authored-by: Woojun Jung <[email protected]> --------- Co-authored-by: Woojun Jung <[email protected]> Co-authored-by: Steven Liu <[email protected]>
1 parent 555693f commit eb4afdd

File tree

2 files changed

+157
-0
lines changed

2 files changed

+157
-0
lines changed

‎docs/source/ko/_toctree.yml

+2
Original file line numberDiff line numberDiff line change
@@ -77,6 +77,8 @@
7777
title: 이미지 특징 추출
7878
- local: tasks/mask_generation
7979
title: 마스크 생성
80+
- local: tasks/keypoint_detection
81+
title: 키포인트 탐지
8082
- local: tasks/knowledge_distillation_for_image_classification
8183
title: 컴퓨터 비전(이미지 분류)를 위한 지식 증류(knowledge distillation)
8284
title: 컴퓨터 비전

‎docs/source/ko/tasks/keypoint_detection.md

+155
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,155 @@
1+
<!--Copyright 2023 The HuggingFace Team. All rights reserved.
2+
3+
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
4+
the License. You may obtain a copy of the License at
5+
6+
http://www.apache.org/licenses/LICENSE-2.0
7+
8+
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
9+
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
10+
specific language governing permissions and limitations under the License.
11+
12+
⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
13+
rendered properly in your Markdown viewer.
14+
15+
-->
16+
17+
# 키포인트 탐지 [[keypoint-detection]]
18+
19+
[[open-in-colab]]
20+
21+
키포인트 감지(Keypoint detection)은 이미지 내의 특정 포인트를 식별하고 위치를 탐지합니다. 이러한 키포인트는 랜드마크라고도 불리며 얼굴 특징이나 물체의 일부와 같은 의미 있는 특징을 나타냅니다.
22+
키포인트 감지 모델들은 이미지를 입력으로 받아 아래와 같은 출력을 반환합니다.
23+
24+
- **키포인트들과 점수**: 관심 포인트들과 해당 포인트에 대한 신뢰도 점수
25+
- **디스크립터(Descriptors)**: 각 키포인트를 둘러싼 이미지 영역의 표현으로 텍스처, 그라데이션, 방향 및 기타 속성을 캡처합니다.
26+
27+
이번 가이드에서는 이미지에서 키포인트를 추출하는 방법을 다루어 보겠습니다.
28+
29+
이번 튜토리얼에서는 키포인트 감지의 기본이 되는 모델인 [SuperPoint](./model_doc/superpoint)를 사용해보겠습니다.
30+
31+
```python
32+
from transformers import AutoImageProcessor, SuperPointForKeypointDetection
33+
processor = AutoImageProcessor.from_pretrained("magic-leap-community/superpoint")
34+
model = SuperPointForKeypointDetection.from_pretrained("magic-leap-community/superpoint")
35+
```
36+
아래의 이미지로 모델을 테스트 해보겠습니다.
37+
38+
<div style="display: flex; align-items: center;">
39+
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"
40+
alt="Bee"
41+
style="height: 200px; object-fit: contain; margin-right: 10px;">
42+
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/cats.png"
43+
alt="Cats"
44+
style="height: 200px; object-fit: contain;">
45+
</div>
46+
47+
48+
```python
49+
import torch
50+
from PIL import Image
51+
import requests
52+
import cv2
53+
54+
55+
url_image_1 = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"
56+
image_1 = Image.open(requests.get(url_image_1, stream=True).raw)
57+
url_image_2 = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/cats.png"
58+
image_2 = Image.open(requests.get(url_image_2, stream=True).raw)
59+
60+
images = [image_1, image_2]
61+
```
62+
63+
이제 입력을 처리하고 추론을 할 수 있습니다.
64+
65+
66+
```python
67+
inputs = processor(images,return_tensors="pt").to(model.device, model.dtype)
68+
outputs = model(**inputs)
69+
```
70+
모델 출력에는 배치 내의 각 항목에 대한 상대적인 키포인트, 디스크립터, 마스크와 점수가 있습니다. 마스크는 이미지에서 키포인트가 있는 영역을 강조하는 역할을 합니다.
71+
72+
```python
73+
SuperPointKeypointDescriptionOutput(loss=None, keypoints=tensor([[[0.0437, 0.0167],
74+
[0.0688, 0.0167],
75+
[0.0172, 0.0188],
76+
...,
77+
[0.5984, 0.9812],
78+
[0.6953, 0.9812]]]),
79+
scores=tensor([[0.0056, 0.0053, 0.0079, ..., 0.0125, 0.0539, 0.0377],
80+
[0.0206, 0.0058, 0.0065, ..., 0.0000, 0.0000, 0.0000]],
81+
grad_fn=<CopySlices>), descriptors=tensor([[[-0.0807, 0.0114, -0.1210, ..., -0.1122, 0.0899, 0.0357],
82+
[-0.0807, 0.0114, -0.1210, ..., -0.1122, 0.0899, 0.0357],
83+
[-0.0807, 0.0114, -0.1210, ..., -0.1122, 0.0899, 0.0357],
84+
...],
85+
grad_fn=<CopySlices>), mask=tensor([[1, 1, 1, ..., 1, 1, 1],
86+
[1, 1, 1, ..., 0, 0, 0]], dtype=torch.int32), hidden_states=None)
87+
```
88+
89+
이미지에 실제 키포인트를 표시하기 위해선 결과값을 후처리 해야합니다. 이를 위해 실제 이미지 크기를 결과값과 함께 `post_process_keypoint_detection`에 전달해야 합니다.
90+
91+
```python
92+
image_sizes = [(image.size[1], image.size[0]) for image in images]
93+
outputs = processor.post_process_keypoint_detection(outputs, image_sizes)
94+
```
95+
96+
위 코드를 통해 결과값은 딕셔너리를 갖는 리스트가 되고, 각 딕셔너리들은 후처리된 키포인트, 점수 및 디스크립터로 이루어져있습니다.
97+
98+
99+
```python
100+
[{'keypoints': tensor([[ 226, 57],
101+
[ 356, 57],
102+
[ 89, 64],
103+
...,
104+
[3604, 3391]], dtype=torch.int32),
105+
'scores': tensor([0.0056, 0.0053, ...], grad_fn=<IndexBackward0>),
106+
'descriptors': tensor([[-0.0807, 0.0114, -0.1210, ..., -0.1122, 0.0899, 0.0357],
107+
[-0.0807, 0.0114, -0.1210, ..., -0.1122, 0.0899, 0.0357]],
108+
grad_fn=<IndexBackward0>)},
109+
{'keypoints': tensor([[ 46, 6],
110+
[ 78, 6],
111+
[422, 6],
112+
[206, 404]], dtype=torch.int32),
113+
'scores': tensor([0.0206, 0.0058, 0.0065, 0.0053, 0.0070, ...,grad_fn=<IndexBackward0>),
114+
'descriptors': tensor([[-0.0525, 0.0726, 0.0270, ..., 0.0389, -0.0189, -0.0211],
115+
[-0.0525, 0.0726, 0.0270, ..., 0.0389, -0.0189, -0.0211]}]
116+
```
117+
118+
이제 위 딕셔너리를 사용하여 키포인트를 표시할 수 있습니다.
119+
120+
```python
121+
import matplotlib.pyplot as plt
122+
import torch
123+
124+
for i in range(len(images)):
125+
keypoints = outputs[i]["keypoints"]
126+
scores = outputs[i]["scores"]
127+
descriptors = outputs[i]["descriptors"]
128+
keypoints = outputs[i]["keypoints"].detach().numpy()
129+
scores = outputs[i]["scores"].detach().numpy()
130+
image = images[i]
131+
image_width, image_height = image.size
132+
133+
plt.axis('off')
134+
plt.imshow(image)
135+
plt.scatter(
136+
keypoints[:, 0],
137+
keypoints[:, 1],
138+
s=scores * 100,
139+
c='cyan',
140+
alpha=0.4
141+
)
142+
plt.show()
143+
```
144+
145+
아래에서 결과를 확인할 수 있습니다.
146+
147+
<div style="display: flex; align-items: center;">
148+
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee_keypoint.png"
149+
alt="Bee"
150+
style="height: 200px; object-fit: contain; margin-right: 10px;">
151+
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/cats_keypoint.png"
152+
alt="Cats"
153+
style="height: 200px; object-fit: contain;">
154+
</div>
155+

0 commit comments

Comments
 (0)