Vision-Language Models (VLMs) demonstrate remarkable capabilities in visual understanding and reasoning, such as in Visual Question Answering (VQA), where the model is asked a question related to a visual input. Still, these models can make distinctly unnatural errors, for example, providing (wrong) answers to unanswerable VQA questions, such as questions asking about objects that do not appear in the image. To address this issue, we propose CLIP-UP: CLIP-based Unanswerable Problem detection, a novel lightweight method for equipping VLMs with the ability to withhold answers to unanswerable questions. CLIP-UP leverages CLIP-based similarity measures to extract question-image alignment information to detect unanswerability, requiring efficient training of only a few additional layers, while keeping the original VLMs' weights unchanged. Tested across several models, CLIP-UP achieves significant improvements on benchmarks assessing unanswerability in both multiple-choice and open-ended VQA, surpassing other methods, while preserving original performance on other tasks.

Balancing content fidelity and artistic style is a pivotal challenge in image generation. While traditional style transfer methods and modern Denoising Diffusion Probabilistic Models (DDPMs) strive to achieve this balance, they often struggle to do so without sacrificing either style, content, or sometimes both. This work addresses this challenge by analyzing the ability of DDPMs to maintain content and style equilibrium. We introduce a novel method to identify sensitivities within the DDPM attention layers, identifying specific layers that correspond to different stylistic aspects. By directing conditional inputs only to these sensitive layers, our approach enables fine-grained control over style and content, significantly reducing issues arising from over-constrained inputs. Our findings demonstrate that this method enhances recent stylization techniques by better aligning style and content, ultimately improving the quality of generated visual content.

Streamlining content discovery within media archives requires integrating advanced data representations and effective visualization techniques for clear communication of video topics to users. The proposed system addresses the challenge of efficiently navigating large video collections by exploiting a fusion of visual, audio, and textual features to accurately index and categorize video content through a text-based method. Additionally, semantic embeddings are employed to provide contextually relevant information and recommendations to users, resulting in an intuitive and engaging exploratory experience over our topics ontology map using OpenAI GPT-4.

Simulations are a powerful tools particularly in the case of safety critical scenarios. However, simulating complex temporal events in multi-agent scenarios with vehicles and pedestrians, such as those that exist in urban environments, is challenging. We present CityLifeSim, a simulation for the research community that focuses on rich pedestrian behavior, such as the one that arises when different personalities, environmental events, and group goals are simulated. In our simulations we can see cases of people jay walking a red light, sitting on a bench, waiting for the bus, or calling on the phone, but also more complex creation and management of crowds that might even line up or just keep moving while observing interpersonal distances. CityLifeSim is configurable and can create unlimited scenarios with detailed logging capabilities. As a demonstration we have run CityLifeSim to create a demo dataset for training setups that includes 17 different cameras, views from a moving vehicle in the street under different weather conditions (rain, snow, sun), and from a drone with frontal and downward views. All content is released with the corresponding original configuration files, ground truth pedestrian segmentation, and RGB-D frames. We evaluate our dataset on a pedestrian detection and identification task with state of the art Multi-Object Tracker (MOT), showing the limitations and opportunities for synthetic data in this use case.

Cartoons and animation domain videos have very different characteristics compared to real-life images and videos. In addition, this domain carries a large variability in styles. Current computer vision and deep-learning solutions often fail on animated content because they were trained on natural images. In this paper we present a method to refine a semantic representation suitable for specific animated content. We first train a neural network on a large-scale set of animation videos and use the mapping to deep features as an embedding space. Next, we use self-supervision to refine the representation for any specific animation style by gathering many examples of animated characters in this style, using a multi-object tracking. These examples are used to define triplets for contrastive loss training. The refined semantic space allows better clustering of animated characters even when they have diverse manifestations. Using this space we can build dictionaries of characters in an animation videos, and define specialized classifiers for specific stylistic content (e.g., characters in a specific animation series) with very little user effort. These classifiers are the basis for automatically labeling characters in animation videos. We present results on a collection of characters in a variety of animation styles. Code and resources are available at: https://github.com/oronnir/CAST .

We developed a generic framework which enables Customer Relationship Management (CRM) organizations to deploy an automated ranking system for leads (commonly known as ‘lead scoring’). Leads are records that represent non-customers who might become customers. Lead ranking is a fundamental CRM problem with many flavors. Ranking serves as a prioritization management tool for CRM organizations, with many characteristics similar to those of recommender systems. We present the system with its most recent developments, emphasizing challenges that go beyond the core of the learning algorithm, and that have played an instrumental role in maturing the system into a trustable feature, robust to different types of organizations and datasets. Particularly, we present features which enable Human in the Loop [1], a dominant concept in both configuration and result consumption. Another type of features demonstrates the addition of domain knowledge into the machine learning based process. We present the concepts of feature selection, with and without human help, prediction explanations, insights on model inputs, data quality issues, training for UX consistency, and actionability for each individual prediction.

Animal dimensions play a vital role in providing data in support of management decisions regarding livestock. Nevertheless, dairy heifers are still measured manually, a time consuming and stressful task for both the farmer and the animal. This research suggests an approach that utilises a fully automated system to measure a heifer's body. The methodology involves a single low-cost Microsoft Kinect V2 Time-of-Flight 3D sensor, computer vision, machine learning, and object recognition using ellipse fitting with quantile regression as part of the feature extraction phase. The camera was installed at the Volcani Center dairy farm, on the ceiling above a free-walk path between the feeding zone and lying area. Video data of 107 Israeli Holstein heifers were recorded and validated against "gold references" (human-observed body mass, hip height and withers height). The tested system improved the normalised Root Mean Squared Error of estimates over the state of the art models by 70.4%, 69.8% and 42.6% for withers height, hip height, and body mass respectively. The models were also validated on a different dairy farm and yielded similar results. The methodology, may be adapted and applied to other elliptically shaped animal bodies, such as sheep, pigs, horses, and buffalo. (C) 2017 lAgrE. Published by Elsevier Ltd. All rights reserved.