Giuliano Grossi

Deterministic 3D hand motion forecasting remains a largely underexplored problem, with substantially less methodological systematization than full-body motion forecasting. Existing hand forecasting approaches are often video-conditioned, multimodal, or probabilistic, whereas spectral motion forecasters typically process retained frequency coefficients with a shared operator. This is restrictive for hand articulation, which combines slow coordinated configuration changes with transient digit-level corrections. We study deterministic sequence-to-sequence forecasting of 3D hand joint trajectories from past landmarks and propose \model{}, a compact frequency-domain forecaster that treats temporal prediction as spectral equalization. The model projects padded input trajectories onto a discrete cosine basis, processes the retained coefficients with parallel MLP branches, and modulates branch responses with learned smooth B-spline gates. The gates are non-exclusive and are applied before branch processing and again at spectral fusion, which allows overlapping frequency specialization without hard routing. A gate-diversity regularizer discourages redundant branch behavior. We evaluate the model on four public datasets, AssemblyHands, H2O, FPHA, and BigHand2.2M, under a unified deterministic protocol over multiple history/horizon ratios. \model{} attains the best result in 17 of 18 AssemblyHands settings, wins all five evaluated settings on H2O, wins four of five on FPHA, and remains competitive on BigHand2.2M, where strong spectral MLP baselines stay close. Ablations indicate that the gains are primarily attributable to this smooth overlapping spectral specialization.

Background: The Critical Assessment of Functional Annotation (CAFA) is a community effort held to understand the field of computational protein function prediction. Every three years, since 2010, the organizers initiate an experiment to collect function predictions on a large set of proteins and then evaluate the performance of predicting methods on a subset of proteins that have accumulated experimental annotations between the submission deadline and the evaluation time. CAFA provides an independent and rigorous assessment of the current state of the art, thus leveling the playing field, highlighting successes, revealing bottlenecks, and offering a forum for the exchange of ideas in protein science. Here, we report the results of the fourth CAFA experiment (CAFA4). Results: CAFA4 featured the participation of 148 methods from 70 research groups on a total of 46,205 unique proteins over a 5-year annotation accumulation phase, the longest in any CAFA. In a comparison across CAFA2-CAFA4 methods, the prediction of Gene Ontology (GO) terms has clearly improved across all three GO aspects and traditional evaluation settings. While not achieving the first rank, several CAFA2 and CAFA3 methods featured in the top ten methods in many evaluations, suggesting that earlier methods still hold relevance. The performance is weaker in the newly introduced “partial knowledge” evaluation category (proteins with experimental annotations before submission deadline that gained additional annotations in the same GO aspect during the annotation accumulation phase), highlighting the need for a new class of methods. The rankings of the methods were stable over the years in traditional evaluation settings, but less so in the new partial knowledge evaluation. Overall, the field continues to progress with some influx of new participants. Sustained efforts will be necessary to substantially advance it.

Remote measurement of respiratory behaviour through RGB cameras has gained significant attention in the last couple of decades. Unlike traditional contact-based methods that may cause discomfort and require specialised equipment, contactless physiological measurement techniques offer a non-invasive way to monitor vital signs. In this survey article, we comprehensively review the literature and techniques related to estimating respiratory information from RGB cameras. We categorise the approaches into three main groups: methods utilising respiration-induced body movements, methods extracting respiratory information from blood volume pulse signals obtained via remote photoplethysmography, and deep learning-based techniques for direct respiratory signal extraction. To evaluate these approaches, we perform a comparative assessment using publicly available datasets. As a result, we uncover emerging trends while identifying strengths and weaknesses in the field. Our contributions include a detailed review of the literature, a benchmark of representative methods on multiple datasets, and the introduction of a new Python package called resPyre that implements the benchmarked approaches, making them accessible to the research community. This survey aims at promoting reproducibility, facilitate further research, and guide the development of more accurate and practical methods for remote respiration measurement using RGB cameras.