Group activity recognition in weakly supervised settings can be approached through either detector-based or detector-free methods. Detector-based methods typically rely on predefined graph structures to capture group relationships, which limits their ability to model the dynamic, multi-scale, and fine-grained changes in group interactions. In contrast, detector-free methods face challenges in effectively filtering irrelevant background and spectator information. In this paper, we propose a novel multi-scale hypergraph neural network framework, MHGAR, for weakly supervised group activity recognition. Unlike traditional methods with fixed graph structures, MHGAR dynamically constructs hyperedges to capture complex team formations and sub-group interactions at multiple scales. We initialize player representations by combining visual appearance and spatial positions, and iteratively refine these representations through dynamic hyperedge construction and bi-directional message passing. A cross-attentive activity classifier integrates the refined multi-scale features with ball position information to predict group activities. Our method effectively captures both fine-grained sub-group interactions and high-level team formations through its adaptive multi-scale modeling. Experimental results demonstrate state-of-the-art performance on the widely used Volleyball and NBA datasets.

Several downscaled datasets of sun-induced chlorophyll fluorescence (SIF) enhance the quality of raw SIF satellite retrievals and offer a better perspective for monitoring terrestrial ecosystems. They are, however, still under pressure in studies that rely on long-term fine-scale observations, such as drought monitoring, climate capture, and gross primary productivity (GPP) estimation, due to persistent temporal and spatial discrimination deficits, spatial gaps, or nonnegligible numerical errors. These limitations are especially noticeable in GOME-2 and its related datasets. This study presented a new machine learning model [called “Geographically random light gradient boosting machine” (GR-LGBM)] and successfully produced a seamless dataset of global daily-mean SIF (GR-LGBM produced SIF) with a spatiotemporal resolution of 0.01° and 8 days for the period 2008–2018, based on GOME-2 SIF, visible-near-infrared reflectance, meteorological, and radiological variables. GR-LGBM performed well in the task of establishing the mapping relationship between the original SIF retrieval and the explanatory variables (fivefold cross-validated${R} ^{2}$: 0.885), outperforming widely used machine learning models such as extreme random forest (EXT), light gradient boosting machine (LGBM), random forest (RF), categorical boosting (CatBoost), Cubist, eXtreme gradient boosting (XGBoost), neural network (NN), and possesses good temporal prediction ability and spatial robustness. The GRSIF001 was highly consistent with SIF satellite retrievals from GOME-2, OCO-2, TROPOMI, and ground-based SIF site observations, demonstrating the spatiotemporal accuracy and temporal scalability of GRSIF001. Additionally, evidence that GRSIF001 precisely captured the actual photosynthesis in vegetation was presented. The downscaling model is better suited for characterizing spatial and temporal heterogeneity of geographic variables, and GRSIF001 will provide greater value in accurately understanding the dynamics of terrestrial photosynthesis.