Amazon Technologies, Inc.

Luitpold Staudigl

Muhammad Bilal Javed

Pablo Barbachano

Thomas Sydney Austin Wallis

Mike Mueller

Michael Donoser

Sabine Sternig

Richard Droste

Martijn Frederik Wouter Van Der Veen

Jochen Gast

Technologies are provided for generating optimized sets of client rendering options for renderable content items (such as images, videos, text, and audio). Rendering optimization parameters for a renderable content item can be determined by assessing the performance of processing requests and rendering content items using different file formats, quality settings, and/or other settings for the renderable content items. The rendering optimization parameters can then be used in combination with a rendering specification that is defined by a developer (such as an application developer, web page designer, or the like) in order to create an optimized set of client rendering options. The set of client rendering options can then be transmitted to one or more client computing devices. A client computing device that receives the set of client rendering options can select one or more of the options in order to render the renderable content item.

Generation of optimized client rendering options

One or more multi-stage optimization iterations are performed with respect to a compression algorithm. A given iteration comprises a first stage in which hyper-parameters of a perceptual quality algorithm are tuned independently of the compression algorithm. A second stage of the iteration comprises tuning hyper-parameters of the compression algorithm using a set of perceptual quality scores generated by the tuned perceptual quality algorithm. The final stage of the iteration comprises performing a compression quality evaluation test on the tuned compression algorithm.

Iterative media object compression algorithm optimization using decoupled calibration of perceptual quality algorithms

Respective labels indicative of compression-related quality degradation for a set of media object tuples which meet a divergence criterion are obtained; each tuple comprises a reference media object and a pair of corresponding compressed media object versions. Pairs of training records for a machine learning model are generated using the labeled media object tuples and multiple perceptual quality algorithms, with each training record comprising respective perceived quality degradation scores generated by each of the multiple algorithms for a given compressed media object of a tuple. A machine learning model is trained, using the record pairs, to predict quality degradation scores for compressed media objects.

Perceived media object quality prediction using adversarial annotations for training and multiple-algorithm scores as input

In operating an ecommerce marketplace, pre-generated metadata is used to transform images after a request is received, in real-time, and with low latency. To accomplish the high-speed image transformation, an offline annotation of relevant image features is obtained using machine learning and metadata is stored based upon the results. The metadata is then used to perform the high-speed image transformation at request time. The transformations can include image cropping, adjustment in saturation, contrast, brightness, extracting portions of the image, etc. In one example, the metadata includes a bounding box giving coordinates of how to crop an image so that the image can be cropped without analyzing content or context of the image. Instead, to obtain the coordinates of the bounding box, content and context of the image are analyzed in pre-request processing. In this way, when the request is received, the more difficult and time-consuming processing is already completed.

Growing community of inventors

Bonn, Germany

Luitpold Staudigl