The design of deep network architectures and training methods in computer vision has been well-explored. However, in almost all cases the images have been used as provided, with little exploration of pre-processing steps beyond normalization and data augmentation. Virtually all images posted on the web or captured by devices are processed for viewing by humans. Is the pipeline used for humans also best for use by computers and deep networks? The human visual system uses logarithmic sensors; differences and sums correspond to ratios and products. Features in log space will be invariant to intensity changes and robust to color balance changes. Log RGB space also reveals structure that is corrupted by typical pre-processing. We explore using linear and log RGB data for training standard backbone architectures on an image classification task using data derived directly from RAW images to guarantee its integrity. We found that networks trained on log RGB data exhibit improved performance on an unmodified test set and invariance to intensity and color balance modifications without additional training or data augmentation. Furthermore, we found that the gains from using high quality log data could also be partially or fully realized from data in 8-bit sRGB-JPG format by inverting the sRGB transform and taking the log. These results imply existing databases may benefit from this type of pre-processing. While working with log data, we found it was critical to retain the integrity of the log relationships and that networks using log data train best with meta-parameters different than those used for sRGB or linear data. Finally, we introduce a new 10-category 10k RAW image data set (RAW10) for image classification and other purposes to enable further the exploration of log RGB as an input format for deep networks in computer vision.

The importance of written communication and critical thinking in Computer Science is widely acknowledged. It was called out specifically in the curriculum guidelines ACM CS2013 [6] and has been the topic of a number of previous SIGCSE papers, for example [1-4]. Moreover, writing as a pedagogical practice can help make CS more accessible for a broader population. However, special challenges may arise for students who are English-language learners or have writing-specific learning differences.

The New York Times published an op-ed by Anita Hill [3] suggesting that women in tech consider class action to remedy the gender bias that is increasingly being reported in the mass-media. This panel raises the question "what are we doing in undergraduate programs to reduce the 'Mad Men', 'Brogrammer' culture she describes that is increasingly being reported in the popular press. Part of our mission as educators is to develop professional behavior so that our students entering the workforce not only understand what it means to act professionally, but understand that it is their responsibility to actively push back on the existing bias within the tech culture. As moderator Ursula Wolz brings a depth of insight from 40 years of industrial and academic experience, including a National Science Foundation project to broaden participation in computing [5]. She does not believe this problem can be solved through quantitative data collection on who does well in computer science, but that SIGCSE needs to begin to collect good stories (ala Sally Fincher [2]) on what constitute best practices to support diversity. The panelists present a range of perspectives that have the potential to establish new cultural norms in the single most influential industry in our economy.

In this special session, we share a variety of ways in which CS 1 courses can broaden their focus beyond programming. These courses strive to effectively introduce the fundamental ideas of computer science, and the importance of computing in broader contexts, without sacrificing programming rigor.

Context-based CS1 courses focusing on Media Computation, Robotics, Games, or Art have been shown to improve outcomes such as retention and gender balance, both important factors in CS education. Colby College has offered a Visual Media focused CS1 course since 2008, and in response to faculty and student feedback, we expanded our curriculum to include a second context-based CS1 course focused on Science applications. Our goal was to have completely different projects but teach the same fundamental concepts. In order to measure whether students in each version were learning the same concepts, and to reduce confounding factors, the same professors co-taught both versions of CS1 and students completed the same homework, quizzes, and final exam. Our analysis of the quiz, final exam, and final overall performance showed no statistically significant difference by context or by gender. There was also no difference by context or gender in whether students took additional CS courses in the following two semesters. Furthermore, as a percentage of the students eligible to take the next offering of CS2, Data Structures and Algorithms, 48% of the students in these two offerings of CS1 registered for CS2, with no significant difference between contexts. Our conclusion is that we were successful in achieving similar outcomes, and the benefits of context-based CS1 courses, in both the Visual Media and Science versions of the course.

Operating robots across networks with unknown, bandwidth, latency and other conditions presents difficulty when the operation depends on real-time feedback and control. Standard video compression methods do a good job compressing arbitrary video, but do not take domain knowledge into account when more information about the video is known beforehand. We have incorporated robot odometry into the video pipeline, allowing video quality to be selectively reduced at times when odometry suggests that such a reduction will not adversely affect task performance of human operators. We found that selectively reducing video quality significantly reduced bandwidth usage, increasing the robot's responsiveness and controllability, while having no measurable effect on task performance.

This paper presents a new model for understanding the appearance of objects that exhibit both body and surface reflection under realistic illumination. Specifically, the model represents the appearance of surfaces that interact with a dominant illuminant and a non-negligible ambient illuminant that may have different spectral power distributions. Real illumination environments usually have an ambient illuminant, and the current dynamic range of consumer cameras is sufficient to capture significant information in shadows. The bi-illuminant dichromatic reflection model explains numerous empirical findings in the literature and has implications for commonly used chromaticity spaces that claim to be illumination invariant but are not in many natural situations. One outcome of the model is the first 2-D chromaticity space for an RGB image that is robust to illumination change given dominant and ambient illuminants with different spectral power distributions.

In this paper, we describe a robot that interacts with humans in a crowded conference environment. The robot detects faces, determines the shirt color of onlooking conference attendants, and reacts with a combination of speech, musical, and movement responses. It continuously updates an internal emotional state, modeled realistically after human psychology research. Using empirically-determined mapping functions, the robot's state in the emotion space is translated to a particular set of sound and movement responses. We successfully demonstrate this system at the AAAI '05 Open Interaction Event, showing the potential for emotional modeling to improve human-robot interaction.

The compass operator has proven to be a useful tool for the detection of color edges in real images. Its fundamental contribution is the comparison of oriented distributions of image features over a local area at each pixel. This paper presents extensions and modifications to the operator to make it applicable to texture edge detection in high dimensional images whose dimensions represent the output of a texture filter bank. The results show that the extended compass operator can robustly locate edges in natural scenes with complex textures. In addition, the use of a dynamic time warping distribution matching metric and jittered application of the operator improves the computational running time by a factor of over 50 while still producing comparable results. This large-scale speedup makes application of the algorithm to an entire image database computationally feasible. 1

This paper provides a survey of the variety of computer vision [CV] and image processing [IP] courses being taught at institutions around the world. The survey shows that, in addition to classic survey courses in CV/IP, there are many focused and multidisciplinary courses being taught that reportedly improve both student and faculty interest in the topic. It also demonstrates that students can successfully undertake a variety of complex lab assignments. In addition, this paper includes a comparative review of current textbooks and supplemental texts appropriate for CV/IP courses.

Computer vision is a broad-based field of computer science that requires students to understand and integrate knowledge from numerous disciplines. Computer science (CS) majors, however, do not necessarily have an interdisciplinary background. In the rush to integrate, we can forget, or fail to plan for the fact that our students may not possess a broad undergraduate education. To explore the appropriateness of our education materials, this paper begins with a discussion of what we can expect CS majors to know and how we can use that knowledge to make a computer vision course a more enriching experience. The paper then provides a review of a number of the currently available computer vision textbooks. These texts differ significantly in their coverage, scope, approach, and audience. This comparative review shows that, while there are an increasing number of good textbooks available, there is still a need for new educational materials. In particular, the field would benefit from both an undergraduate computer vision text aimed at computer scientists and from a text with a stronger focus on color computer vision and its applications.

We previously presented a framework for segmentation of complex scenes using multiple physical hypotheses for simple image regions. A consequence of that framework was a proposal for a new approach to the segmentation of complex scenes into regions corresponding to coherent surfaces rather than merely regions of similar color. Herein we present an implementation of this new approach and show example segmentations for scenes containing multi-colored piece-wise uniform objects. Using our approach we are able to intelligently segment scenes with objects of greater complexity than previous physics-based segmentation algorithms. The results show that by using general physical models we obtain segmentations that correspond more closely to coherent surfaces in the scene than segmentations found using only color.

This paper presents a new approach to segmentation using explicit hypotheses about the physics that creates images. We propose an initial segmentation that identifies image regions exhibiting constant color; but possibly varying intensity. For each region, we propose a set of hypotheses, each of which specifically models the illumination, reflectance, and shape of the 3-D patch which caused that region. Each hypothesis represents a distinct, plausible explanation for the color and intensity variation of that patch. Hypotheses for adjacent patches can be compared for similarity and merged when appropriate, resulting in more global hypotheses which group elementary regions.>