Amazon.com

About: Amazon.com is a company organization based out in Seattle, Washington, United States. It is known for research contribution in the topics: Computer science & Service (business). The organization has 13363 authors who have published 17317 publications receiving 266589 citations.

Box2Seg: Attention weighted loss and discriminative feature learning for weakly supervised segmentation

Abstract: We propose a weakly supervised approach to semantic segmentation using bounding box annotations. Bounding boxes are treated as noisy labels for the foreground objects. We predict a per-class attention map that saliently guides the per-pixel cross entropy loss to focus on foreground pixels and refines the segmentation boundaries. This avoids propagating erroneous gradients due to incorrect foreground labels on the background. Additionally, we learn pixel embeddings to simultaneously optimize for high intra-class feature affinity while increasing discrimination between features across different classes. Our method, Box2Seg, achieves state-of-the-art segmentation accuracy on PASCAL VOC 2012 by significantly improving the mIOU metric by \(2.1\%\) compared to previous weakly supervised approaches. Our weakly supervised approach is comparable to the recent fully supervised methods when fine-tuned with limited amount of pixel-level annotations. Qualitative results and ablation studies show the benefit of different loss terms on the overall performance.

Mediation of wakeword response for multiple devices

Abstract: Devices, systems, and methods provide synchronization of multiple voice-controlled devices to establish priority of one of the devices to respond to an acoustic signal, preventing other devices from responding to a single user command. Each device is configured to mediate its response to a wakeword by, after detecting the wakeword in an audio input, generating a data packet and attempting to send it to the other devices over a communication channel that prevents data collision. One device will succeed in sending the packet, while the other devices cannot until they receive the first device's packet, indicating that the first device has priority. Additionally, the devices may process their audio inputs to determine a signal quality, the signal qualities are shared between all devices, and the device with the best signal quality assumes priority.

Conditions influencing the adoption of effective anti-deforestation policies in South America’s commodity frontiers

Abstract: Reducing large-scale deforestation in commodity frontiers remains a key challenge for climate change mitigation and the conservation of biodiversity. Public and private anti-deforestation policies have been shown to effectively reduce forest loss, but the conditions under which such policies get adopted are rarely examined. Here we propose a set of conditions that we expect to be associated with the adoption of effective anti-deforestation policies in commodity frontiers. We then examine whether these conditions have influenced policy adoption in South America’s major soy-and-cattle frontiers: the Brazilian Amazon, the Cerrado, the Chaco, the Chiquitano, and Paraguay’s Atlantic Forest. By collating empirical data from diverse sources, including literature review, extensive expert interviews, and analysis of primary and secondary data, we show that the Cerrado, the Chaco, and the Chiquitano differ from the Brazilian Amazon in multiple ways that might have inhibited adoptions of effective anti-deforestation instruments. These conditions include: a higher importance of the agricultural sector within the respective countries, lower carbon stocks and species richness, higher prevalence of private land tenure, and higher baseline compliance with forest reserve regulations on private lands. We also observe that the adoption of the most effective private anti-deforestation instrument, commodity moratoria, may respond to similar conditions as those influencing the adoption of public instruments. Incentivizing public and private actors to adopt effective anti-deforestation policies in the Cerrado, Chaco, and the Chiquitano will likely be more challenging than it has been in the Brazilian Amazon.

System for resolving ambiguous queries based on user context

Abstract: A system for resolving a user's question regarding media at the user's location where the question includes ambiguity as to the context of the question. Context information is acquired from a device or devices at the user's location, and if multiple possible contexts are identified, the system may attempt to resolve the question for more than one of the possible contexts. The system for answering questions and the source of the media may be different and independent.

Providing files with cacheable portions

Abstract: Systems and methods are provided for enabling caching of one or more portions of a file. In some embodiments, a first file may be sent to a client system, where the first file includes a cache identifier identifying a shared portion of content of the first file, where the shared portion of content is included in both the first file and at least one other file. A request may be received from the client system for a second file that includes the shared portion of content. A modified second page may then be sent to the client system, where the modified second page includes the cache identifier, and does not include the shared portion of content, such that the client system may determine the content of the second page based on the modified second page and the previously stored shared portion of content.