This paper strives to recognize individual actions and group activities from videos. While existing solutions for this challenging problem explicitly model spatial and temporal relationships based on location of individual actors, we propose an actor-transformer model able to learn and selectively extract information relevant for group activity recognition. We feed the transformer with rich actorspecific static and dynamic representations expressed by features from a 2D pose network and 3D CNN, respectively. We empirically study different ways to combine these representations and show their complementary benefits. Experiments show what is important to transform and how itshould be transformed. What is more, actor-transformers achieve state-of-the-art results on two publicly available benchmarks for group activity recognition, outperforming the previous best published results by a considerable margin.

IMPORTANCE Despite the introduction of combination antiretroviral therapy (cART), HIV-associated neurocognitive disorders continue to be a problem for treated HIV-positive individuals. The cause of this impairment remains unclear.OBJECTIVE To determine if detectable brain changes occur during a 2-year period in HIV-positive individuals who were aviremic and treated with cART. DESIGN, SETTING, AND PARTICIPANTSIn this longitudinal case-control study, participants underwent neuroimaging and neuropsychological assessment approximately 2 years apart. Data were collected from October 26, 2011, to March 1, 2016. Data from 92 HIV-positive individuals were acquired at Washington University in St Louis from ongoing studies conducted in the infectious disease clinic and AIDS Clinical Trial Unit. A total of 55 HIV-negative control participants were recruited from the St Louis community and a research participant registry. A total of 48 HIV-positive individuals who were aviremic and treated with cART and 31 demographically similar HIV-negative controls met the study requirements and were included in the analyses.MAIN OUTCOMES AND MEASURES Brain volumes were extracted with tensor-based and voxel-based morphometry and cortical modeling. Raw scores from neuropsychological tests quantified cognitive performance. Multivariable mixed-effects models assessed the effect of HIV serostatus on brain volumes and cognitive performance, and determined if HIV serostatus affected how these measures changed over time. With HIV-positive participants, linear regression models tested whether brain volumes and cognitive performance were associated with measures of infection severity and duration of infection.

Background Cognitive impairment still occurs in a substantial subset of HIV-infected patients, despite effective viral suppression with highly active antiretroviral therapy (HAART). Structural brain changes may provide clues about the underlying pathophysiology. This study provides a detailed spatial characterization of the pattern and extent of brain volume changes associated with HIV, and relates these brain measures to cognitive ability and clinical variables. Methods Multiple novel neuroimaging techniques (deformation-based morphometry, voxel-based morphometry and cortical modeling) were used to assess regional brain volumes in 125 HIV-infected patients and 62 HIV-uninfected individuals. 90% of the HIV-infected patients were on stable HAART with a majority (75%) having plasma viral suppression. Brain volumetrics and cortical thickness estimates were compared between the HIV-infected and uninfected groups, and the relationships between these measures of brain volume and indices of current and past infection severity, central nervous system penetration of HAART, and cognitive performance were assessed. Results Regionally specific patterns of reduced thalamic and brainstem volumes, as well as reduced cortical thickness in the orbitofrontal cortex, cingulate gyrus, primary motor and sensory cortex, temporal, and frontal lobes were seen in HIV-infected patients compared to HIV-uninfected participants. Observed white matter loss and subcortical atrophy were associated with lower nadir CD4 cell counts, while reduction in cortical thickness was related to worse cognitive performance. Conclusion Our findings suggest that distinct mechanisms may underlie cortical and subcortical injury in people with HIV, and argues for the potential importance of early initiation of HAART to protect long term brain health.