open_images_v4

ओपन इमेज ~9M छवियों का एक डेटासेट है जिसे इमेज-लेवल लेबल और ऑब्जेक्ट बाउंडिंग बॉक्स के साथ एनोटेट किया गया है।

V4 के प्रशिक्षण सेट में 1.74M छवियों पर 600 ऑब्जेक्ट वर्गों के लिए 14.6M बाउंडिंग बॉक्स हैं, जो इसे ऑब्जेक्ट स्थान एनोटेशन के साथ सबसे बड़ा मौजूदा डेटासेट बनाता है। सटीकता और स्थिरता सुनिश्चित करने के लिए बक्से बड़े पैमाने पर पेशेवर एनोटेटर द्वारा मैन्युअल रूप से तैयार किए गए हैं। छवियां बहुत विविध हैं और अक्सर कई वस्तुओं के साथ जटिल दृश्य होते हैं (8.4 प्रति छवि औसतन)। इसके अलावा, डेटासेट को हजारों वर्गों में फैले छवि-स्तरीय लेबल के साथ एनोटेट किया गया है।

विभाजित करना उदाहरण
'test' 125,436
'train' 1,743,042
'validation' 41,620
  • फ़ीचर संरचना :
FeaturesDict({
    'bobjects': Sequence({
        'bbox': BBoxFeature(shape=(4,), dtype=float32),
        'is_depiction': int8,
        'is_group_of': int8,
        'is_inside': int8,
        'is_occluded': int8,
        'is_truncated': int8,
        'label': ClassLabel(shape=(), dtype=int64, num_classes=601),
        'source': ClassLabel(shape=(), dtype=int64, num_classes=6),
    }),
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'image/filename': Text(shape=(), dtype=string),
    'objects': Sequence({
        'confidence': int32,
        'label': ClassLabel(shape=(), dtype=int64, num_classes=19995),
        'source': ClassLabel(shape=(), dtype=int64, num_classes=6),
    }),
    'objects_trainable': Sequence({
        'confidence': int32,
        'label': ClassLabel(shape=(), dtype=int64, num_classes=7186),
        'source': ClassLabel(shape=(), dtype=int64, num_classes=6),
    }),
})
  • फ़ीचर दस्तावेज़ीकरण :
विशेषता कक्षा आकार डीटाइप विवरण
विशेषताएं डिक्ट
bojects क्रम
बॉबजेक्ट्स/बॉक्स बीबॉक्स फीचर (4,) फ्लोट32
bobjects/is_depiction टेन्सर int8
बॉबजेक्ट्स/is_group_of टेन्सर int8
बॉबजेक्ट्स/is_inside टेन्सर int8
ऑब्जेक्ट/is_occluded टेन्सर int8
bobjects/is_truncated टेन्सर int8
बॉबजेक्ट्स/लेबल क्लासलेबल int64
विषय/स्रोत क्लासलेबल int64
छवि छवि (कोई नहीं, कोई नहीं, 3) uint8
छवि/फ़ाइल नाम मूलपाठ डोरी
वस्तुओं क्रम
वस्तुओं/आत्मविश्वास टेन्सर int32
ऑब्जेक्ट्स / लेबल क्लासलेबल int64
ऑब्जेक्ट्स/स्रोत क्लासलेबल int64
object_trainable क्रम
ऑब्जेक्ट्स_ट्रेनेबल / कॉन्फिडेंस टेन्सर int32
ऑब्जेक्ट्स_ट्रेनेबल/लेबल क्लासलेबल int64
ऑब्जेक्ट्स_ट्रेनेबल/स्रोत क्लासलेबल int64
  • पर्यवेक्षित कुंजियाँ ( as_supervised doc देखें): None

  • उद्धरण :

@article{OpenImages,
  author = {Alina Kuznetsova and
            Hassan Rom and
            Neil Alldrin and
            Jasper Uijlings and
            Ivan Krasin and
            Jordi Pont-Tuset and
            Shahab Kamali and
            Stefan Popov and
            Matteo Malloci and
            Tom Duerig and
            Vittorio Ferrari},
  title = {The Open Images Dataset V4: Unified image classification,
           object detection, and visual relationship detection at scale},
  year = {2018},
  journal = {arXiv:1811.00982}
}
@article{OpenImages2,
  author = {Krasin, Ivan and
            Duerig, Tom and
            Alldrin, Neil and
            Ferrari, Vittorio
            and Abu-El-Haija, Sami and
            Kuznetsova, Alina and
            Rom, Hassan and
            Uijlings, Jasper and
            Popov, Stefan and
            Kamali, Shahab and
            Malloci, Matteo and
            Pont-Tuset, Jordi and
            Veit, Andreas and
            Belongie, Serge and
            Gomes, Victor and
            Gupta, Abhinav and
            Sun, Chen and
            Chechik, Gal and
            Cai, David and
            Feng, Zheyun and
            Narayanan, Dhyanesh and
            Murphy, Kevin},
  title = {OpenImages: A public dataset for large-scale multi-label and
           multi-class image classification.},
  journal = {Dataset available from
             https://storage.googleapis.com/openimages/web/index.html},
  year={2017}
}

open_images_v4/मूल (डिफ़ॉल्ट कॉन्फ़िगरेशन)

  • कॉन्फ़िग विवरण : इमेज उनके मूल रिज़ॉल्यूशन और गुणवत्ता पर।

  • डेटासेट का आकार : 562.42 GiB

  • चित्र ( tfds.show_examples ):

VISUALIZATION

open_images_v4/300k

  • कॉन्फ़िग विवरण : छवियों में लगभग 300,000 पिक्सेल हैं, 72 JPEG गुणवत्ता पर।

  • डेटासेट का आकार : 81.92 GiB

  • चित्र ( tfds.show_examples ):

VISUALIZATION

open_images_v4/200k

  • कॉन्फिग विवरण : छवियों में लगभग 200,000 पिक्सेल हैं, 72 JPEG गुणवत्ता पर।

  • डेटासेट का आकार : 60.70 GiB

  • चित्र ( tfds.show_examples ):

VISUALIZATION