An official website of the United States government
Exploiting relationships between objects for image and video captioning has received increasing attention. Most existing methods depend heavily on pre-trained detectors of objects and their relationships, and thus may not work well when facing detection challenges such as heavy occlusion, tiny-size objects, and long-tail classes. In this paper, we propose a joint commonsense and relation reasoning method that exploits prior knowledge for image and video captioning without relying on any detectors. The prior knowledge provides semantic correlations and constraints between objects, serving as guidance to build semantic graphs that summarize object relationships, some of which cannot be directly perceived from images or videos. Particularly, our method is implemented by an iterative learning algorithm that alternates between 1) commonsense reasoning for embedding visual regions into the semantic space to build a semantic graph and 2) relation reasoning for encoding semantic graphs to generate sentences. Experiments on several benchmark datasets validate the effectiveness of our prior knowledge-based approach.
more »
« less
Video Relationship Reasoning using Gated Spatio-Temporal Energy Graph
Visual relationship reasoning is a crucial yet challenging task for understanding rich interactions across visual concepts. For example, a relationship 'man, open, door' involves a complex relation 'open' between concrete entities 'man, door'. While much of the existing work has studied this problem in the context of still images, understanding visual relationships in videos has received limited attention. Due to their temporal nature, videos enable us to model and reason about a more comprehensive set of visual relationships, such as those requiring multiple (temporal) observations (e.g., 'man, lift up, box' vs. 'man, put down, box'), as well as relationships that are often correlated through time (e.g., 'woman, pay, money' followed by 'woman, buy, coffee'). In this paper, we construct a Conditional Random Field on a fully-connected spatio-temporal graph that exploits the statistical dependency between relational entities spatially and temporally. We introduce a novel gated energy function parametrization that learns adaptive relations conditioned on visual observations. Our model optimization is computationally efficient, and its space computation complexity is significantly amortized through our proposed parameterization. Experimental results on benchmark video datasets (ImageNet Video and Charades) demonstrate state-of-the-art performance across three standard relationship reasoning tasks: Detection, Tagging, and Recognition.
more »
« less
- Award ID(s):
- 1750439
- PAR ID:
- 10099440
- Date Published:
- Journal Name:
- Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Video entailment aims at determining if a hypothesis textual statement is entailed or contradicted by a premise video. The main challenge of video entailment is that it requires fine-grained reasoning to understand the complex and long story-based videos. To this end, we propose to incorporate visual grounding to the entailment by explicitly linking the entities described in the statement to the evidence in the video. If the entities are grounded in the video, we enhance the entailment judgment by focusing on the frames where the entities occur. Besides, in the entailment dataset, the entailed/contradictory (also named as real/fake) statements are formed in pairs with subtle discrepancy, which allows an add-on explanation module to predict which words or phrases make the statement contradictory to the video and regularize the training of the entailment judgment. Experimental results demonstrate that our approach outperforms the state-of-the-art methods.more » « less
-
In this research, we take an innovative approach to the Video Corpus Visual Answer Localization (VCVAL) task using the MedVidQA dataset. We expand on it by incorporating causal inference for medical videos, a novel approach in this field. By leveraging the state-of-the-art GPT-4 and Gemini Pro 1.5 models, the system aims to localize temporal segments in videos and analyze cause-effect relationships from subtitles to enhance medical decision-making. This paper extends the work from the MedVidQA challenge by introducing causality extraction to enhance the interpretability of localized video content. Subtitles are segmented to identify causal units such as cause, effect, condition, action, and signal. Prompts guide GPT-4 and Gemini Pro 1.5 in detecting and quantifying causal structures while analyzing explicit and implicit relationships, including those spanning multiple subtitle fragments. Our results reveal that both GPT-4 and Gemini Pro 1.5 perform better when handling queries individually but face challenges in batch processing for both temporal localization and causality extraction. Despite these challenges, our innovative approach has the potential to significantly advance the field of Health Informatics. In this research, we address the Video Corpus Visual Answer Localization (VCVAL) task using the MedVidQA dataset and take it a step further by integrating causal inference for medical videos. By leveraging the state-of-the-art GPT-4 and Gemini Pro 1.5 model, our system is designed to localize temporal segments in videos and analyze cause-effect relationships from subtitles to enhance medical decision-making. Our preliminary results indicate that while both models perform well for some videos, they face challenges for most, resulting in varying performance levels. The successful integration of temporal localization with causal inference can provide significant improvement for the scalability and overall performance of medical video analysis. Our work demonstrates how AI systems can uncover valuable insights from medical videos, driving significant progress in medical AI applications and potentially making significant contributions to the field.more » « less
-
As intelligent systems increasingly blend into our everyday life, artificial social intelligence becomes a prominent area of research. Intelligent systems must be socially intelligent in order to comprehend human intents and maintain a rich level of interaction with humans. Human language offers a unique unconstrained approach to probe through questions and reason through answers about social situations. This unconstrained approach extends previous attempts to model social intelligence through numeric supervision (e.g. sentiment and emotions labels). In this paper, we introduce the Social-IQ, an unconstrained benchmark specifically designed to train and evaluate socially intelligent technologies. By providing a rich source of open-ended questions and answers, Social-IQ opens the door to explainable social intelligence. The dataset contains rigorously annotated and validated videos, questions and answers, as well as annotations for the complexity level of each question and answer. Social- IQ contains 1, 250 natural in-thewild social situations, 7, 500 questions and 52, 500 correct and incorrect answers. Although humans can reason about social situations with very high accuracy (95.08%), existing state-of-the-art computational models struggle on this task. As a result, Social-IQ brings novel challenges that will spark future research in social intelligence modeling, visual reasoning, and multimodal question answering (QA).more » « less
-
Introduction This dataset was gathered during the Vid2Real online video-based study, which investigates humans’ perception of robots' intelligence in the context of an incidental Human-Robot encounter. The dataset contains participants' questionnaire responses to four video study conditions, namely Baseline, Verbal, Body language, and Body language + Verbal. The videos depict a scenario where a pedestrian incidentally encounters a quadruped robot trying to enter a building. The robot uses verbal commands or body language to try to ask for help from the pedestrian in different study conditions. The differences in the conditions were manipulated using the robot’s verbal and expressive movement functionalities. Dataset Purpose The dataset includes the responses of human subjects about the robots' social intelligence used to validate the hypothesis that robot social intelligence is positively correlated with human compliance in an incidental human-robot encounter context. The video based dataset was also developed to obtain empirical evidence that can be used to design future real-world HRI studies. Dataset Contents Four videos, each corresponding to a study condition. Four sets of Perceived Social Intelligence Scale data. Each set corresponds to one study condition Four sets of compliance likelihood questions, each set include one Likert question and one free-form question One set of Godspeed questionnaire data. One set of Anthropomorphism questionnaire data. A csv file containing the participants demographic data, Likert scale data, and text responses. A data dictionary explaining the meaning of each of the fields in the csv file. Study Conditions There are 4 videos (i.e. study conditions), the video scenarios are as follows. Baseline: The robot walks up to the entrance and waits for the pedestrian to open the door without any additional behaviors. This is also the "control" condition. Verbal: The robot walks up to the entrance, and says ”can you please open the door for me” to the pedestrian while facing the same direction, then waits for the pedestrian to open the door. Body Language: The robot walks up to the entrance, turns its head to look at the pedestrian, then turns its head to face the door, and waits for the pedestrian to open the door. Body Language + Verbal: The robot walks up to the entrance, turns its head to look at the pedestrian, and says ”Can you open the door for me” to the pedestrian, then waits for the pedestrian to open the door. Image showing the Verbal condition. Image showing the Body Language condition. A within-subject design was adopted, and all participants experienced all conditions. The order of the videos, as well as the PSI scales, were randomized. After receiving consent from the participants, they were presented with one video, followed by the PSI questions and the two exploratory questions (compliance likelihood) described above. This set was repeated 4 times, after which the participants would answer their general perceptions of the robot with Godspeed and AMPH questionnaires. Each video was around 20 seconds and the total study time was around 10 minutes. Video as a Study Method A video-based study in human-robot interaction research is a common method for data collection. Videos can easily be distributed via online participant recruiting platforms, and can reach a larger sample than in-person/lab-based studies. Therefore, it is a fast and easy method for data collection for research aiming to obtain empirical evidence. Video Filming The videos were filmed with a first-person point-of-view in order to maximize the alignment of video and real-world settings. The device used for the recording was an iPhone 12 pro, and the videos were shot in 4k 60 fps. For better accessibility, the videos have been converted to lower resolutions. Instruments The questionnaires used in the study include the Perceived Social Intelligence Scale (PSI), Godspeed Questionnaire, and Anthropomorphism Questionnaire (AMPH). In addition to these questionnaires, a 5-point Likert question and a free-text response measuring human compliance were added for the purpose of the video-based study. Participant demographic data was also collected. Questionnaire items are attached as part of this dataset. Human Subjects For the purpose of this project, the participants are recruited through Prolific. Therefore, the participants are users of Prolific. Additionally, they are restricted to people who are currently living in the United States, fluent in English, and have no hearing or visual impairments. No other restrictions were imposed. Among the 385 participants, 194 participants identified as female, and 191 as male, the age ranged from 19 to 75 (M = 38.53, SD = 12.86). Human subjects remained anonymous. Participants were compensated with $4 upon submission approval. This study was reviewed and approved by UT Austin Internal Review Board. Robot The dataset contains data about humans’ perceived social intelligence of a Boston Dynamics’ quadruped robot Spot (Explorer model). The robot was selected because quadruped robots are gradually being adopted to provide services such as delivery, surveillance, and rescue. However, there are still issues or obstacles that robots cannot easily overcome by themselves in which they will have to ask for help from nearby humans. Therefore, it is important to understand how humans react to a quadruped robot that they incidentally encounter. For the purposes of this video-study, the robot operation was semi-autonomous, with the navigation being manually teleoperated by an operator and a few standalone autonomous modules to supplement it. Data Collection The data was collected through Qualtrics, a survey development platform. After the completion of data collection, the data was downloaded as a csv file. Data Quality Control Qualtrics automatically detects bots so any response that is flagged as bots are discarded. All incomplete and duplicate responses were discarded. Data Usage This dataset can be used to conduct a meta-analysis on robots' perceived intelligence. Please note that data is coupled with this study design. Users interested in data reuse will have to assess that this dataset is in line with their study design. Acknowledgement This study was funded through the NSF Award # 2219236GCR: Community-Embedded Robotics: Understanding Sociotechnical Interactions with Long-term Autonomous Deployments.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
