Crowdsourcing image analysis for plant phenomics to generate ground truth data for machine learning.

The accuracy of machine learning tasks critically depends on high quality ground truth data. Therefore, in many cases, producing good ground truth data typically involves trained professionals; however, this can be costly in time, effort, and money. Here we explore the use of crowdsourcing to generate a large number of training data of good quality. We explore an image analysis task involving the segmentation of corn tassels from images taken in a field setting. We investigate the accuracy, speed and other quality metrics when this task is performed by students for academic credit, Amazon MTurk workers, and Master Amazon MTurk workers. We conclude that the Amazon MTurk and Master Mturk workers perform significantly better than the for-credit students, but with no significant difference between the two MTurk worker types. Furthermore, the quality of the segmentation produced by Amazon MTurk workers rivals that of an expert worker. We provide best practices to assess the quality of ground truth data, and to compare data quality produced by different sources. We conclude that properly managed crowdsourcing can be used to establish large volumes of viable ground truth data at a low cost and high quality, especially in the context of high throughput plant phenotyping. We also provide several metrics for assessing the quality of the generated datasets.

Selection of macroreference frames in spatial memory.

Spatial memories are often hierarchically organized with different regions of space represented in unique clusters within the hierarchy. Each cluster is thought to be organized around its own microreference frame selected during learning, whereas relationships between clusters are organized by a macroreference frame. Two experiments were conducted in order to better understand important characteristics of macroreference frames. Participants learned overlapping spatial layouts of objects within a room-sized environment before performing a perspective-taking task from memory. Of critical importance were between-layout judgments thought to reflect the macroreference frame. The results indicate that (1) macroreference frames characterize overlapping spatial layouts, (2) macroreference frames are used even when microreference frames are aligned with one another, and (3) macroreference frame selection depends on an interaction between the global macroaxis (defined by characteristics of the layout of all learned objects), the relational macroaxis (defined by characteristics of the two layouts being related on a perspective-taking trial), and the learning view. These results refine the current understanding of macroreference frames and document their broad role in spatial memory.

Rescaling of perceived space transfers across virtual environments.

Research over the past 20 years has consistently shown that egocentric distance is underperceived in virtual environments (VEs) compared with real environments. In 2 experiments, judgments of object distance (Experiment 1) and object size (Experiment 2) improved after a brief period of walking through the VE with continuous visual feedback. Whereas improvement of blind-walking distance judgments could be attributable to recalibration of walking, improvement in perceived size is considered evidence for rescaling of perceived space, whereby perceived size and distance increased after walking interaction. Furthermore, improvements in judged distance and size transferred to a new VE. Distance judgments, but not size judgments, continued to improve after additional walking interaction in the new VE. These results have theoretical implications regarding the effects of walking interaction on perceived space, and practical implications regarding methods of improving perceived distance in VEs. (PsycINFO Database Record

Walking through a virtual environment improves perceived size within and beyond the walked space.

Distances tend to be underperceived in virtual environments (VEs) by up to 50%, whereas distances tend to be perceived accurately in the real world. Previous work has shown that allowing participants to interact with the VE while receiving continual visual feedback can reduce this underperception. Judgments of virtual object size have been used to measure whether this improvement is due to the rescaling of perceived space, but there is disagreement within the literature as to whether judgments of object size benefit from interaction with feedback. This study contributes to that discussion by employing a more natural measure of object size. We also examined whether any improvement in virtual distance perception was limited to the space used for interaction (1-5 m) or extended beyond (7-11 m). The results indicated that object size judgments do benefit from interaction with the VE, and that this benefit extends to distances beyond the explored space.

Frontal extents in virtual environments are not immune to underperception.

Distance is commonly underperceived by up to 50 % in virtual environments (VEs), in contrast to relatively accurate real world judgments. Experiments reported by Geuss, Stefanucci, Creem-Regehr, and Thompson (Journal of Experimental Psychology: Human Perception and Performance, 38, 1242-1253, 2012) indicate that the exocentric distance separating two objects in a VE is underperceived when the objects are oriented in the sagittal plane (depth extents), but veridically perceived when oriented in a frontoparallel plane (frontal extents). The authors conclude that "distance underestimation in the [VE] generalizes to intervals in the depth plane, but not to intervals in the frontal plane." The current experiment evaluated an alternative hypothesis that the accurate judgments of frontal extents reported by Geuss et al. were due to a fortunate balance of underperception caused by the VE and overperception of frontal relative to depth extents. Participants judged frontal and depth extents in the classroom VE used by Geuss et al. and in a sparser VE containing only a grass-covered ground plane. Judgments in the classroom VE replicated findings by Geuss et al., but judgments in the grass VE show underperception of both depth and frontal extents, indicating that frontal extents are not immune to underperception in VEs.

Recalibration of perceived distance in virtual environments occurs rapidly and transfers asymmetrically across scale.

Distance in immersive virtual reality is commonly underperceived relative to intended distance, causing virtual environments to appear smaller than they actually are. However, a brief period of interaction by walking through the virtual environment with visual feedback can cause dramatic improvement in perceived distance. The goal of the current project was to determine how quickly improvement occurs as a result of walking interaction (Experiment 1) and whether improvement is specific to the distances experienced during interaction, or whether improvement transfers across scales of space (Experiment 2). The results show that five interaction trials resulted in a large improvement in perceived distance, and that subsequent walking interactions showed continued but diminished improvement. Furthermore, interaction with near objects (1-2 m) improved distance perception for near but not far (4-5 m) objects, whereas interaction with far objects broadly improved distance perception for both near and far objects. These results have practical implications for ameliorating distance underperception in immersive virtual reality, as well as theoretical implications for distinguishing between theories of how walking interaction influences perceived distance.