Massive free visual data sets for machine learning and deep learning
Image analysis is the extraction of data from a digital image using digital image processing techniques. It involves processing an image into fundamental components in order to extract statistical data. It includes tasks such as finding shapes, detecting edges, counting objects, forensic, biometrics, face recognition etc.
The human visual cortex is an excellent image analyser, especially for extracting higher-level information. Human vision can’t be replaced by computers in analysing an image. Many important image analysis tools such as edge detectors and neural networks are inspired by human visual perception models.
Many different techniques are used to automatically analyse images. Each technique may be useful for a small range of tasks. Examples of image analysis techniques include 2D and 3D object recognition, motion detection, video tracking, Automatic number plate recognition etc.
We can consider a simple example of face recognition used in Facebook.
Face recognition is a computer application, capable of identifying a person from a digital image or a video source. There are many ways to identify or recognise the face. Facebook compares selected features from the image and database (the profile picture) and compares them to suggest whom to tag.
Video Content Analysis
Video content analysis (VCA) is analysing the video to detect and determine events not based on just a single image. Examples in different fields include Dynamic masking, Flame and smoke detection, Motion detection, shape recognition, object detection, video tracking etc.
Example for Video content analysis is the Closed-circuit television (CCTV) also known as video surveillance. It uses the video camera to transmit signals to a specific place, on a limited set of monitors. It captures all the moving objects.
Open data sets available
In order to fine-tune algorithms that recognise and predict patterns, you need to feed massive amounts of previously-tagged information to test and learn. You can always count on Google to have data, generated by the users who interact with and upload content to its services. Google uses that data to build intelligence for the company. Google has uploaded a massive amount of data for machine learning and deep learning last year.
The Open Images Dataset
The Open Images Dataset comes from a collaboration between Google, Carnegie Mellon and Cornell, with 9 million entries that were tagged by computers first before having those notes verified and corrected by humans. The Google Research team says it has enough images to train a neural network “from scratch” so if you’d like to try your hand at a DeepDream-style project, better version of Google Photos or the next Prisma then it’s ready to go. Plus, plans are underway to improve the quality of the annotations in Open Images the coming months.
The YouTube-8M Dataset is compiled from 8 million YouTube videos (adding up to more than 500,000 hours of footage). It aims for diversity and quality. You can explore the data set online or download it, but the data set is only available in the TensorFlow Record file format. The idea is to create a library for video analysis that rivals already existing still images, which are accessible for people without big data. Part of that is because Google has also extracted and tagged still images from the videos for researchers to download.
So that the machine learning community can make use of all these data and build better computational algorithms.