Emilian Radoi

Accurately predicting vehicle attributes from an image is a challenging yet crucial task with numerous applications in the automotive industry, including car identification, price estimation, and traffic monitoring. Traditional Computer Vision approaches have primarily focused on single-objective tasks where the model either classifies a single feature of the vehicle or predicts a class that aggregates multiple characteristics. These methods often overlook the inter-dependencies between different vehicle attributes and may not fully capture the fine-grained separation of information that Deep Learning models should learn. To address these limitations, we propose partitioning vehicle characteristics into a hierarchical structure where information at an upper level (supertask) conditions the information at the lower level (subtask). To leverage this hierarchical structure, we propose a multi-task neural network architecture that conditions the predictions of subtasks based on its prediction for the supertask using attention and gating mechanisms. This conditioning is further reinforced by a proposed task-consistency loss which is designed to teach the model the task hierarchy in scenarios where the training data is limited. We evaluate our approach on multiple image-based vehicle classification benchmarks, including our proposed CarModelDataset, a difficult dataset specifically collected for a comprehensive analysis. Our proposed methodology and architecture obtain an F1 score of 97.14% on Stanford Cars, 85.08% on Autovit, and 63.03% on CarModelDataset. Our results show that the hierarchical conditioning approach outperforms single-task models in all scenarios, especially in contexts with limited data, highlighting the effectiveness of our method. We release the code^a and our best performing model for research purposes, and make the CarModelDataset available upon request.

Gait encodes rich biometric and behavioural information, yet leveraging the manner of walking to infer psychological traits remains a challenging and underexplored problem. We introduce a hierarchical Multi-Stage Mixture of Movement Experts (MoME) architecture for multi-task prediction of psychological attributes from gait sequences represented as 2D poses. MoME processes the walking cycle in four stages of movement complexity, employing lightweight expert models to extract spatio-temporal features and task-specific gating modules to adaptively weight experts across traits and stages. Evaluated on the PsyMo benchmark covering 17 psychological traits, our method outperforms state-of-the-art gait analysis models, achieving a 37.47% weighted F1 score at the run level and 44.6% at the subject level. Our experiments show that integrating auxiliary tasks such as identity recognition, gender prediction, and BMI estimation further improves psychological trait estimation. Our findings demonstrate the viability of multi-task gait-based learning for psychological trait estimation and provide a foundation for future research on movement-informed psychological inference.

Gait recognition has emerged as a powerful tool for unobtrusive and long-range identity analysis, with growing relevance in surveillance and monitoring applications. Although recent advances in deep learning and large-scale datasets have enabled highly accurate recognition under closed-set conditions, real-world deployment demands open-set gait enrollment, which means determining whether a new gait sample corresponds to a known identity or represents a previously unseen individual. In this work, we introduce a transformer-based framework for open-set gait enrollment that is both dataset-agnostic and recognition-architecture-agnostic. Our method leverages a SetTransformer to make enrollment decisions based on the embedding of a probe sample and a context set drawn from the gallery, without requiring task-specific thresholds or retraining for new environments. By decoupling enrollment from the main recognition pipeline, our model is generalized across different datasets, gallery sizes, and identity distributions. We propose an evaluation protocol that uses existing datasets in different ratios of identities and walks per identity. We instantiate our method using skeleton-based gait representations and evaluate it on two benchmark datasets (CASIA-B and PsyMo), using embeddings from three state-of-the-art recognition models (GaitGraph, GaitFormer, and GaitPT). We show that our method is flexible, is able to accurately perform enrollment in different scenarios, and scales better with data compared to traditional approaches. We will make the code and dataset scenarios publicly available.

Gait recognition from video streams is a challenging problem in computer vision biometrics due to the subtle differences between gaits and numerous confounding factors. Recent advancements in self-supervised pretraining have led to the development of robust gait recognition models that are invariant to walking covariates. While neural scaling laws have transformed model development in other domains by linking performance to data, model size, and compute, their applicability to gait remains unexplored. In this work, we conduct the first empirical study scaling on skeleton-based self-supervised gait recognition to quantify the effect of data quantity, model size and compute on downstream gait recognition performance. We pretrain multiple variants of GaitPT - a transformer-based architecture - on a dataset of 2.7 million walking sequences collected in the wild. We evaluate zero-shot performance across four benchmark datasets to derive scaling laws for data, model size, and compute. Our findings demonstrate predictable power-law improvements in performance with increased scale and confirm that data and compute scaling significantly influence downstream accuracy. We further isolate architectural contributions by comparing GaitPT with GaitFormer under controlled compute budgets. These results provide practical insights into resource allocation and performance estimation for real-world gait recognition systems.

Andrei Niculae, Adrian Cosma, Cosmin Dumitrache, Emilian Radoi. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track. 2025.

Andrei Niculae, Adrian Cosma, Emilian Radoi. Proceedings of the 24th Workshop on Biomedical Language Processing. 2025.

Large Language Models (LLMs) have demonstrated remarkable capabilities in various domains, including data augmentation and synthetic data generation. This work explores the use of LLMs to generate rich textual descriptions for motion sequences, encompassing both actions and walking patterns. We leverage the expressive power of LLMs to align motion representations with high-level linguistic cues, addressing two distinct tasks: action recognition and retrieval of walking sequences based on appearance attributes. For action recognition, we employ LLMs to generate textual descriptions of actions in the BABEL-60 dataset, facilitating the alignment of motion sequences with linguistic representations. In the domain of gait analysis, we investigate the impact of appearance attributes on walking patterns by generating textual descriptions of motion sequences from the DenseGait dataset using LLMs. These descriptions capture subtle variations in walking styles influenced by factors such as clothing choices and footwear. Our approach demonstrates the potential of LLMs in augmenting structured motion attributes and aligning multi-modal representations. The findings contribute to the advancement of comprehensive motion understanding and open up new avenues for leveraging LLMs in multi-modal alignment and data augmentation for motion analysis. We make the code publicly available at https://github.com/Radu1999/WalkAndText

Surveillance footage represents a valuable resource and opportunities for conducting gait analysis. However, the typical low quality and high noise levels in such footage can severely impact the accuracy of pose estimation algorithms, which are foundational for reliable gait analysis. Existing literature suggests a direct correlation between the efficacy of pose estimation and the subsequent gait analysis results. A common mitigation strategy involves fine-tuning pose estimation models on noisy data to improve robustness. However, this approach may degrade the downstream model's performance on the original high-quality data, leading to a trade-off that is undesirable in practice. We propose a processing pipeline that incorporates a task-targeted artifact correction model specifically designed to pre-process and enhance surveillance footage before pose estimation. Our artifact correction model is optimized to work alongside a state-of-the-art pose estimation network, HRNet, without requiring repeated fine-tuning of the pose estimation model. Furthermore, we propose a simple and robust method for obtaining low quality videos that are annotated with poses in an automatic manner with the purpose of training the artifact correction model. We systematically evaluate the performance of our artifact correction model against a range of noisy surveillance data and demonstrate that our approach not only achieves improved pose estimation on low-quality surveillance footage, but also preserves the integrity of the pose estimation on high resolution footage. Our experiments show a clear enhancement in gait analysis performance, supporting the viability of the proposed method as a superior alternative to direct fine-tuning strategies. Our contributions pave the way for more reliable gait analysis using surveillance data in real-world applications, regardless of data quality.

Gait, an unobtrusive biometric, is valued for its capability to identify individuals at a distance, across external outfits and environmental conditions. This study challenges the prevailing assumption that vision-based gait recognition, in particular skeleton-based gait recognition, relies primarily on motion patterns, revealing a significant role of the implicit anthropometric information encoded in the walking sequence. We show through a comparative analysis that removing height information leads to notable performance degradation across three models and two benchmarks (CASIA-B and GREW). Furthermore, we propose a spatial transformer model processing individual poses, disregarding any temporal information, which achieves unreasonably good accuracy, emphasizing the bias towards appearance information and indicating spurious correlations in existing benchmarks. These findings underscore the need for a nuanced understanding of the interplay between motion and appearance in vision-based gait recognition, prompting a reevaluation of the methodological assumptions in this field. Our experiments indicate that "in-the-wild" datasets are less prone to spurious correlations, prompting the need for more diverse and large scale datasets for advancing the field.

Gaze estimation, the task of predicting where an individual is looking, is a critical task with direct applications in areas such as human-computer interaction and virtual reality. Estimating the direction of looking in unconstrained environments is difficult, due to the many factors that can obscure the face and eye regions. In this work we propose CrossGaze, a strong baseline for gaze estimation, that leverages recent developments in computer vision architectures and attention-based modules. Unlike previous approaches, our method does not require a specialised architecture, utilizing already established models that we integrate in our architecture and adapt for the task of 3D gaze estimation. This approach allows for seamless updates to the architecture as any module can be replaced with more powerful feature extractors. On the Gaze360 benchmark, our model surpasses several state-of-the-art methods, achieving a mean angular error of 9.94 degrees. Our proposed model serves as a strong foundation for future research and development in gaze estimation, paving the way for practical and accurate gaze prediction in real-world scenarios.