Google Releases Big Dataset of Images with Annotations

[lpetrich]

Google AI Blog: Announcing Open Images V4 and the ECCV 2018 Open Images Challenge

Today, we are happy to announce Open Images V4, containing 15.4M bounding-boxes for 600 categories on 1.9M images, making it the largest existing dataset with object location annotations. The boxes have been largely manually drawn by professional annotators to ensure accuracy and consistency. The images are very diverse and often contain complex scenes with several objects (8 per image on average; visualizer).

The annotations are boxes on the images that denote something recognized.

Here it is: Open Images Dataset V4 The dataset comes with a challenge: Open Images Challenge 2018

1. Object Class Detection: predicting a tight bounding box around all instances of the 500 classes.
2. Visual Relationship Detection: detecting pairs of objects in particular relations, e.g. "woman playing guitar".

Object recognition in images has some rather big challenges:

• The size of the data. A 300*300 image contains nearly 100,000 pixels, each with 3 color channels.
• Different sizes, orientations, and lightings
• Different features of the objects themselves, like different colors
• Only part of an object being visible, the rest being obstructed by foreground objects and/or outside the boundaries of the image file

As to why Google might be doing something so seemingly altruistic, it may be to use universities and the like as a farm team -- whoever does well could get some nice job at Google.

 

[Kharakov]

Google image search still sucks at fractals.