Computer Vision Researcher for Novel Video Causal Reasoning Benchmark

We are looking for an experienced Computer Vision / Video Understanding researcher to help develop a benchmark model for a new video dataset and annotation scheme. This is a research-oriented project. We are specifically looking for someone who can understand a novel dataset formulation, translate it into a well-defined modeling problem, and implement an appropriate novel model architecture, rather than simply fine-tuning an existing model. [What You’ll Work On] Our dataset represents causal relationships between events in videos as graphs. The goal is to develop a model that takes video/event information as input and predicts the causal graph among events. You will work with us to: Understand our new dataset and causal-relation annotation scheme Formulate the corresponding benchmark/modeling task Design a graph-based architecture for video causal reasoning Implement and train the proposed model Develop appropriate baselines and ablation studies Evaluate the model using rigorous benchmark metrics Help ensure that the experimental design is suitable for a top-tier computer vision / machine learning publication Relevant existing work includes MECD, MECD+, and MM-OR. Familiarity with video causal reasoning, graph prediction, temporal reasoning, or video-language models is highly desirable. [Required Qualifications] We are looking for a researcher with: Strong expertise in computer vision, video understanding, or multimodal learning Experience with graph-based neural architectures, causal/temporal reasoning, or structured prediction Strong implementation experience with PyTorch or equivalent frameworks Experience designing and running rigorous research experiments Ability to independently understand research papers and translate new research ideas into working model architectures First-author publication experience at a top-tier computer vision / machine learning conference, such as CVPR, ICCV, ECCV, NeurIPS, ICML, or ICLR Experience with video-language models, Graph Neural Networks (GNNs), Transformers, or causal reasoning benchmarks is a strong plus. Most Relevant Prior Work The closest existing models/datasets to our intended direction include: MECD / MECD+ — video event causal discovery and causal reasoning MM-OR — structured multimodal understanding and graph prediction in operating-room environments Our goal is not simply to reproduce these models. We want to use them as starting points while developing an architecture specifically suited to our new causal graph annotation scheme and benchmark formulation. [When Applying] Please include: "Links or PDFs of your first-author publications" A brief description of your most relevant experience in video understanding, causal reasoning, graph-based modeling, or multimodal learning Your role in designing and implementing the models in those publications GitHub or other code samples, if available Please apply only if you have substantial research experience developing novel deep-learning architectures. This project is particularly suitable for a researcher with strong academic publication experience who is comfortable working on an open-ended research problem, rather than a standard model implementation or engineering task.