Résultats de recherche pour « Temporal Smoothing »

Abstract Online action detection (OAD) is a challenging task that involves predicting the ongoing action class in real-time from streaming videos. In this article, we propose a novel approach for OAD based on the RWKV model and introducing temporal label smoothing. The RWKV model effectively captures temporal dependencies while maintaining computational efficiency, making it well-suited for real-time applications. The temporal label smoothing technique enhances the model's robustness in handling ambiguous boundaries and feature similarities near action boundaries. We conducted experiments on two widely used datasets, THUMOS'14 and TVSeries, to evaluate the performance of our proposed approach. Our model achieves state-of-the-art performance with 71.8% mAP on THUMOS'14 and 89.7% cAP on TVSeries. Furthermore, our proposed approach demonstrates impressive efficiency, running at over 600 FPS and maintaining a competitive mAP of 59.9% on THUMOS'14 with RGB features alone. This high efficiency makes our model suitable for real-time deployment, even on resource-constrained devices. These results showcase the effectiveness and competitiveness of our proposed approach in OAD.

Abstract Surgical phase recognition from endoscopic video could enable numerous context-aware technologies that impact efficiency and performance of surgeons and minimally invasive care teams. However, surgical phases can vary greatly (from seconds to minutes) due to patient factors and surgeon workflows along with many other reasons. Given this wide range of phase durations, fixed temporal parameters of neural networks and machine learning models constrain surgical phases with different temporal rhythms. To address this problem, we ensemble neural networks and other machine learning models with different architectures and temporal parameters to recognize surgical phases. The probability estimates of the ensemble are then used as priors for forward-backward smoothing to generate posteriors of surgical phases. We demonstrate the performance of this modeling process on three data sets: 1) robot-assisted inguinal hernia (five phases), 2) robot-assisted training in a porcine model (seven phases), and 3) laparoscopic cholecystectomy (Cholec80, seven phases). The results suggest that this novel method to address varying phases in different procedures holds promise for the future of surgical phase recognition.

Abstract We propose an approximation to the forward filter backward sampler (FFBS) algorithm for large‐scale spatio‐temporal smoothing. FFBS is commonly used in Bayesian statistics when working with linear Gaussian state‐space models, but it requires inverting covariance matrices which have the size of the latent state vector. The computational burden associated with this operation effectively prohibits its applications in high‐dimensional settings. We propose a scalable spatio‐temporal FFBS approach based on the hierarchical Vecchia approximation of Gaussian processes, which has been previously successfully used in spatial statistics. On simulated and real data, our approach outperformed a low‐rank FFBS approximation.

Facilitated by advancements in conflict event databases, studies have moved towards predicting armed conflict and understanding its determinants subnationally. However, existing statistical models do not capture the diffusion of armed conflict, and hence do not adequately account for its dependence across both time and space, thus inducing potential biases in the estimates of covariates of interest. To address this, we introduce a regression approach that simultaneously captures both spatial and temporal dimension of the diffusion of armed conflict through non-parametric smoothing. Using fine-grained conflict data on Africa, we demonstrate the importance of accounting for this dependence, capturing conflict diffusion up to 550km in distance and 24 months in the past, and observe that diffusion decays exponentially in both space and time. We illustrate the flexibility of our fully interpretable method by studying the role of population in the transmission of conflict. We find that conflict typically breaks out in densely populated areas, and from there diffuses, specifically to lower population areas.