No abstract

Semantic understanding of visual scenes is one of the holy grails of computer vision. Despite efforts of the community in data collection, there are still few image datasets covering a wide range of scenes and object categories with pixel-wise annotations for scene understanding. In this work, we present a densely annotated dataset ADE20K, which spans diverse annotations of scenes, objects, parts of objects, and in some cases even parts of parts. Totally there are 25k images of the complex everyday scenes containing a variety of objects in their natural spatial context. On average there are 19.5 instances and 10.5 object classes per image. Based on ADE20K, we construct benchmarks for scene parsing and instance segmentation. We provide baseline performances on both of the benchmarks and re-implement the state-ofthe-art models for open source. We further evaluate the effect of synchronized batch normalization and find that a reasonably large batch size is crucial for the semantic segmentation performance. We show that the networks trained on ADE20K are able to segment a wide variety of scenes and objects 1 .

In this paper, we are interested in modeling complex activities that occur in a typical household. We propose to use programs, i.e., sequences of atomic actions and interactions, as a high level representation of complex tasks. Programs are interesting because they provide a non-ambiguous representation of a task, and allow agents to execute them. However, nowadays, there is no database providing this type of information. Towards this goal, we first crowd-source programs for a variety of activities that happen in people's homes, via a game-like interface used for teaching kids how to code. Using the collected dataset, we show how we can learn to extract programs directly from natural language descriptions or from videos. We then implement the most common atomic (inter)actions in the Unity3D game engine, and use our programs to "drive" an artificial agent to execute tasks in a simulated household environment. Our VirtualHome simulator allows us to create a large activity video dataset with rich ground-truth, enabling training and testing of video understanding models. We further showcase examples of our agent performing tasks in our VirtualHome based on language descriptions.

Background The banning of mass-gathering indoor events to prevent SARS-CoV-2 spread has had an important effect on local economies. Despite growing evidence on the suitability of antigen-detecting rapid diagnostic tests (Ag-RDT) for mass screening at the event entry, this strategy has not been assessed under controlled conditions. We aimed to assess the effectiveness of a prevention strategy during a live indoor concert. MethodsWe designed a randomised controlled open-label trial to assess the effectiveness of a comprehensive preventive intervention for a mass-gathering indoor event (a live concert) based on systematic same-day screening of attendees with Ag-RDTs, use of facial masks, and adequate air ventilation. The event took place in the Sala Apolo, Barcelona, Spain. Adults aged 18-59 years with a negative result in an Ag-RDT from a nasopharyngeal swab collected immediately before entering the event were randomised 1:1 (block randomisation stratified by age and gender) to either attend the indoor event for 5 hours or go home. Nasopharyngeal specimens used for Ag-RDT screening were analysed by real-time reverse-transcriptase PCR (RT-PCR) and cell culture (Vero E6 cells). 8 days after the event, a nasopharyngeal swab was collected and analysed by Ag-RDT, RT-PCR, and a transcription-mediated amplification test (TMA). The primary outcome was the difference in incidence of RT-PCR-confirmed SARS-CoV-2 infection at 8 days between the control and the intervention groups, assessed in all participants who were randomly assigned, attended the event, and had a valid result for the SARS-CoV-2 test done at follow-up. The trial is registered at ClinicalTrials.gov, NCT04668625. Findings Participant enrollment took place during the morning of the day of the concert, Dec 12, 2020. Of the 1140 people who responded to the call and were deemed eligible, 1047 were randomly assigned to either enter the music event (experimental group) or continue with normal life (control group). Of the 523 randomly assigned to the experimental group, 465 were included in the analysis of the primary outcome (51 did not enter the event and eight did not take part in the follow-up assessment), and of the 524 randomly assigned to the control group, 495 were included in the final analysis (29 did not take part in the follow-up). At baseline, 15 (3%) of 495 individuals in the control group and 13 (3%) of 465 in the experimental group tested positive on TMA despite a negative Ag-RDT result. The RT-PCR test was positive in one case in each group and cell viral culture was negative in all cases. 8 days after the event, two (<1%) individuals in the control arm had a positive Ag-RDT and PCR result, whereas no Ag-RDT nor RT-PCR positive results were found in the intervention arm. The Bayesian estimate for the incidence between the experimental and control groups was -0•15% (95% CI -0•72 to 0•44). Interpretation Our study provides preliminary evidence on the safety of indoor mass-gathering events during a COVID-19 outbreak under a comprehensive preventive inte...

No abstract