In this paper, we describe the design and evaluation of the toolkit Bitacora, addressed to practitioners working in non-profit organizations interested in integrating Twitter data into their work. The toolkit responds to the call to maintain the locality of data by promoting a qualitative and contextualized approach to analyzing Twitter data. We assessed the toolkit’s effectiveness in guiding practitioners to search, collect, and be critical when analyzing data from Twitter. We evaluated the toolkit with ten practitioners from three non-profit organizations of different aims and sizes in Mexico. The assessment surfaced tensions between the assumptions embedded in the toolkit’s design and practitioners’ expectations, needs, and backgrounds. We show that practitioners navigated these tensions in some cases by developing strategies and, in others, questioning the appropriateness of using Twitter data to inform their work. We conclude with recommendations for researchers who developed tools for non-profit organizations to inform humanitarian action.

This paper examines the practices involved in mobilizing social media data from their site of production to the institutional context of non-profit organizations. We report on nine months of fieldwork with a transnational and intergovernmental organization using social media data to understand the role of grassroots initiatives in Mexico, in the unique context of the COVID-19 pandemic. We show how different stakeholders negotiate the definition of problems to be addressed with social media data, the collective creation of ground-truth, and the limitations involved in the process of extracting value from data. The meanings of social media data are not defined in advance; instead, they are contingent on the practices and needs of the organization that seeks to extract insights from the analysis. We conclude with a list of reflections and questions for researchers who mediate in the mobilization of social media data into non-profit organizations to inform humanitarian action.

The peculiar nature of whole slide imaging (WSI), digitizing conventional glass slides to obtain multiple high resolution images which capture microscopic details of a patient’s histopathological features, has garnered increased interest from the computer vision research community over the last two decades. Given the unique computational space and time complexity inherent to gigapixel-size whole slide image data, researchers have proposed novel machine learning algorithms to aid in the performance of diagnostic tasks in clinical pathology. One effective algorithm represents a Whole slide image as a bag of smaller image patches, which can be represented as low-dimension image patch embeddings. Weakly supervised deep-learning methods, such as cluster-constrained-attention multiple instance learning (CLAM), have shown promising results when combined with image patch embeddings. While traditional ensemble classifiers yield improved task performance, such methods come with a steep cost in model complexity. Through knowledge distillation, it is possible to retain some performance improvements from an ensemble, while minimizing costs to model complexity. In this work, we implement a weakly supervised ensemble using clustering-constrained-attention multiple-instance learners (CLAM), which uses attention and instance-level clustering to identify task salient regions and feature extraction in whole slides. By applying logit-based and attention-based knowledge distillation, we show it is possible to retain some performance improvements resulting from the ensemble at zero cost to model complexity.

The ability to track simulated cancer cells through the circulatory system, important for developing a mechanistic understanding of metastatic spread, pushes the limits of today's supercomputers by requiring the simulation of large fluid volumes at cellular-scale resolution. To overcome this challenge, we introduce a new adaptive physics refinement (APR) method that captures cellular-scale interaction across large domains and leverages a hybrid CPU-GPU approach to maximize performance. Through algorithmic advances that integrate multi-physics and multi-resolution models, we establish a finely resolved window with explicitly modeled cells coupled to a coarsely resolved bulk fluid domain. In this work we present multiple validations of the APR framework by comparing against fully resolved fluid-structure interaction methods and employ techniques, such as latency hiding and maximizing memory bandwidth, to effectively utilize heterogeneous node architectures. Collectively, these computational developments and performance optimizations provide a robust and scalable framework to enable system-level simulations of cancer cell transport.

Large-scale crowd simulation and visualization is crucial for the next generation of interactive virtual environments. Current authoring techniques produce good results but are laborious, and demand valuable graphics memory and computational resources beyond the reach of consumer-level hardware. In this paper, we propose a technique for generating animatable characters for crowds that reduces memory requirements. The first step consists in reducing, segmenting and labeling a data set of virtual characters into simpler body parts; labeling information is then manually generated and used to correctly match different body parts in order to generate new characters. The second step comprises a method to embed the rig and skinning information into the texture space shared among the new characters. Additional methods using color, skin features, pattern, fat, wrinkle and textile fold maps are used to add more variety. Animation sequences are stored in auxiliary textures. These can be transferred between different characters, as well as versions of the same characters with different level of detail; such animations can be modified, and otherwise reused, increasing variety yet reducing memory requirements. We will demonstrate that our technique has four advantages: first, memory requirements are reduced by 91% when compared to traditional libraries; second, it can also generate previously nonexistent characters from the original data set; third, embedding the rig and skinning into texture space allows painless animation transfer between different characters and fourth, between different levels of detail of the same characters.

Populated virtual environments need to be simulated with as much variety as possible. By identifying the most salient parts of the scene and characters, available resources can be concentrated where they are needed most. In this paper, we investigate which body parts of virtual characters are most looked at in scenes containing duplicate characters or clones . Using an eye-tracking device, we recorded fixations on body parts while participants were asked to indicate whether clones were present or not. We found that the head and upper torso attract the majority of first fixations in a scene and are attended to most. This is true regardless of the orientation, presence or absence of motion, sex, age, size, and clothing style of the character. We developed a selective variation method to exploit this knowledge and perceptually validated our method. We found that selective colour variation is as effective at generating the illusion of variety as full colour variation. We then evaluated the effectiveness of four variation methods that varied only salient parts of the characters. We found that head accessories, top texture and face texture variation are all equally effective at creating variety, whereas facial geometry alterations are less so. Performance implications and guidelines are presented.

Adding hair to virtual humans has focused on three important aspects: hair modeling, hair animation and hair rendering. We present an easy-to-use tool for hairstyling, modeling and animation of hair in real time. For modeling we use a paint-based interface to specify geometrical and styling characteristics of hair while for animation we use a simplified physical based approximation.

Corner measurement is of main concern within the following tasks: camera calibration, image matching, object tracking, recognition and reconstruction. This paper presents a hybrid evolutionary ridge regression approach for the problem of corner modeling. We search model parameters characterizing L-corner models by means of fitting the model to the image data. As the model fitting relies on an initial parameter estimation, we use a global approach to find the global minimum. Experimental results applied to an L-corner using several levels of noise show the advantages and disadvantages of our evolutionary algorithm compared to down-hill simplex and simulated annealing.