COCO 數據集使用說明書

時間 2019-11-24

標籤 coco 數據使用說明書简体版

原文原文鏈接

下面的代碼改寫自 COCO 官方 API，改寫後的代碼 cocoz.py 被我放置在 Xinering/cocoapi。個人主要改進有：html

增長對 Windows 系統的支持；
替換 defaultdict 爲 dict.get()，解決 Windows 的編碼問題。
跳過解壓這一步驟（包括直接的或間接的解壓），直接對圖片數據 images 與標註數據 annotations 操做。
由於，無需解壓，因此 API 的使用更加便捷和高效。

具體的 API 使用說明見以下內容：python

0 準備

COCOZ 簡介git

爲了可使用 cocoz，你須要下載 Xinering/cocoapi。以後將其放在你須要運行的項目或程序根目錄，亦或者使用以下命令添加環境變量（暫時的）：github

import sys
sys.path.append('D:\API\cocoapi\PythonAPI')  # 你下載的 cocoapi 所在路徑

from pycocotools.cocoz import AnnZ, ImageZ, COCOZ   # 載入 cocoz

下面咱們就能夠利用這個 API 的 cocoz.AnnZ、cocoz.ImageZ 和 cocoz.COCOZ 類來操做 COCO 圖片和標註了。下面我以 Windows 系統爲例說明，Linux 是相似的。json

1 cocoz.AnnZ 與 cocoz.ImageZ

root = r'E:\Data\coco'   # COCO 數據集所在根目錄
annType = 'annotations_trainval2017'   # COCO 標註數據類型

annZ = AnnZ(root, annType)

咱們來查看一下，該標註數據所包含的標註類型：api

annZ.names

['annotations/instances_train2017.json',
 'annotations/instances_val2017.json',
 'annotations/captions_train2017.json',
 'annotations/captions_val2017.json',
 'annotations/person_keypoints_train2017.json',
 'annotations/person_keypoints_val2017.json']

以 dict 的形式載入 'annotations/instances_train2017.json' 的具體信息：網絡

annFile = 'annotations/instances_val2017.json'
dataset = annZ.json2dict(annFile)

Loading json in memory ...
used time: 0.890035 s

dataset.keys()

dict_keys(['info', 'licenses', 'images', 'annotations', 'categories'])

dataset['images'][0]  # 記錄了一張圖片的一些標註信息

{'license': 4,
 'file_name': '000000397133.jpg',
 'coco_url': 'http://images.cocodataset.org/val2017/000000397133.jpg',
 'height': 427,
 'width': 640,
 'date_captured': '2013-11-14 17:02:52',
 'flickr_url': 'http://farm7.staticflickr.com/6116/6255196340_da26cf2c9e_z.jpg',
 'id': 397133}

1.1 從網頁獲取圖片

%pylab inline
import skimage.io as sio

coco_url = dataset['images'][0]['coco_url']
# use url to load image
I = sio.imread(coco_url)
plt.axis('off')
plt.imshow(I)
plt.show()

Populating the interactive namespace from numpy and matplotlib

1.2 從本地讀取圖片

爲了不解壓數據集，我使用了 zipfile 模塊：app

imgType = 'val2017'
imgZ = ImageZ(root, imgType)

I = imgZ.buffer2array(imgZ.names[0])

plt.axis('off')
plt.imshow(I)
plt.show()

2 cocoz.COCOZ

root = r'E:\Data\coco'   # COCO 數據集所在根目錄
annType = 'annotations_trainval2017'   # COCO 標註數據類型
annFile = 'annotations/instances_val2017.json'

annZ = AnnZ(root, annType)
coco = COCOZ(annZ, annFile)

Loading json in memory ...
used time: 1.02004 s
Loading json in memory ...
creating index...
index created!
used time: 0.431003 s

若是你須要預覽你載入的 COCO 數據集，可使用 print() 來實現：dom

print(coco)

description: COCO 2017 Dataset
url: http://cocodataset.org
version: 1.0
year: 2017
contributor: COCO Consortium
date_created: 2017/09/01

coco.keys()

dict_keys(['dataset', 'anns', 'imgToAnns', 'catToImgs', 'imgs', 'cats'])

2.1 展現 COCO 的類別與超類

cats = coco.loadCats(coco.getCatIds())
nms = set([cat['name'] for cat in cats])  # 獲取 cat 的 name 信息
print('COCO categories: \n{}\n'.format(' '.join(nms)))
# ============================================================
snms = set([cat['supercategory'] for cat in cats])  # 獲取 cat 的 name 信息
print('COCO supercategories: \n{}'.format(' '.join(snms)))

COCO categories: 
kite potted plant handbag clock umbrella sports ball bird frisbee toilet toaster spoon car snowboard banana fire hydrant skis chair tv skateboard wine glass tie cell phone cake zebra baseball glove stop sign airplane bed surfboard cup knife apple broccoli bicycle train carrot remote cat bear teddy bear person bench horse dog couch orange hair drier backpack giraffe sandwich book donut sink oven refrigerator boat mouse laptop toothbrush keyboard truck motorcycle bottle pizza traffic light cow microwave scissors bus baseball bat elephant fork bowl tennis racket suitcase vase sheep parking meter dining table hot dog

COCO supercategories: 
accessory furniture sports vehicle appliance electronic animal indoor outdoor person kitchen food

2.2 經過給定條件獲取圖片

獲取包含給定類別的全部圖片electron

# get all images containing given categories, select one at random
catIds = coco.getCatIds(catNms=['cat', 'dog', 'snowboar'])  # 獲取 Cat 的 Ids
imgIds = coco.getImgIds(catIds=catIds )  # 
img = coco.loadImgs(imgIds)

隨機選擇一張圖片的信息：

img = coco.loadImgs(imgIds[np.random.randint(0,len(imgIds))])[0]

img

{'license': 4,
 'file_name': '000000318238.jpg',
 'coco_url': 'http://images.cocodataset.org/val2017/000000318238.jpg',
 'height': 640,
 'width': 478,
 'date_captured': '2013-11-21 00:01:06',
 'flickr_url': 'http://farm8.staticflickr.com/7402/9964003514_84ce7550c9_z.jpg',
 'id': 318238}

2.2.1 獲取圖片

從網絡獲取圖片：

coco_url = img['coco_url']

I = sio.imread(coco_url)
plt.axis('off')
plt.imshow(I)
plt.show()

從本地獲取圖片：

這裏有一個梗：cv2 的圖片默認模式是 BGR 而不是 RGB,因此，將 I 直接使用 plt 會改變原圖的顏色空間，爲此咱們可使用 cv2.COLOR_BGR2RGB.

imgType = 'val2017'
imgZ = ImageZ(root, imgType)

I = imgZ.buffer2array(img['file_name'])

plt.axis('off')
plt.imshow(I)
plt.show()

2.3 將圖片的 anns 信息標註在圖片上

# load and display instance annotations
plt.imshow(I)
plt.axis('off')
annIds = coco.getAnnIds(imgIds=img['id'], catIds=catIds, iscrowd=None)
anns = coco.loadAnns(annIds)
coco.showAnns(anns)

2.4 關鍵點檢測

# initialize COCO api for person keypoints annotations
root = r'E:\Data\coco'   # COCO 數據集所在根目錄
annType = 'annotations_trainval2017'   # COCO 標註數據類型
annFile = 'annotations/person_keypoints_val2017.json'

annZ = AnnZ(root, annType)
coco_kps = COCOZ(annZ, annFile)

Loading json in memory ...
used time: 0.882997 s
Loading json in memory ...
creating index...
index created!
used time: 0.368036 s

先選擇一張帶有 person 的圖片：

catIds = coco.getCatIds(catNms=['person'])  # 獲取 Cat 的 Ids
imgIds = coco.getImgIds(catIds=catIds)  
img = coco.loadImgs(imgIds)[77]

# use url to load image
I = sio.imread(img['coco_url'])
plt.axis('off')
plt.imshow(I)
plt.show()

# load and display keypoints annotations
plt.imshow(I); plt.axis('off')
ax = plt.gca()
annIds = coco_kps.getAnnIds(imgIds=img['id'], catIds=catIds, iscrowd=None)
anns = coco_kps.loadAnns(annIds)
coco_kps.showAnns(anns)

2.5 看圖說話

# initialize COCO api for person keypoints annotations
root = r'E:\Data\coco'   # COCO 數據集所在根目錄
annType = 'annotations_trainval2017'   # COCO 標註數據類型
annFile = 'annotations/captions_val2017.json'

annZ = AnnZ(root, annType)
coco_caps = COCOZ(annZ, annFile)

Loading json in memory ...
used time: 0.435748 s
Loading json in memory ...
creating index...
index created!
used time: 0.0139964 s

# load and display caption annotations
annIds = coco_caps.getAnnIds(imgIds=img['id']);
anns = coco_caps.loadAnns(annIds)
coco_caps.showAnns(anns)
plt.imshow(I)
plt.axis('off')
plt.show()

show：

A brown horse standing next to a woman in front of a house.
a person standing next to a horse next to a building
A woman stands beside a large brown horse.
The woman stands next to the large brown horse.
A woman hold a brown horse while a woman watches.

若是你須要使用官方 API, 能夠參考 COCO 數據集的使用。

若是你以爲對你有幫助，請幫忙在 Github 上點個 star：datasetsome。該教程的代碼我放在了 GitHub: COCOZ 使用說明書。