Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
EventMG: Efficient Multilevel Mamba-Graph Learning for Spatiotemporal Event Representation
Authors: Sheng Wu, Lin Jin, Hui Feng, Bo Hu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental evaluation on representative dynamic tasks indicates that Event MG delivers performance comparable to state-of-the-art heavyweight models, but at a lower computational and parametric cost. The results underscore the potential of our proposed architectural paradigm, which seeks to strike a deliberate balance between computational efficiency, representational power, and respect for the intrinsic properties of event data. |
| Researcher Affiliation | Academia | Sheng Wu1 Lin Jin1 Hui Feng1,2 Bo Hu1,2 1 College of Future Information Technology, Fudan University 2 State Key Laboratory of Integrated Chips and Systems, Fudan University |
| Pseudocode | No | The paper describes the methodology and architecture in detail but does not include any explicitly labeled pseudocode or algorithm blocks in a structured format. |
| Open Source Code | No | Due to time constraints, the experimental code for this version does not yet meet the standard for public release, but we plan to make it publicly available in the future. |
| Open Datasets | Yes | Object Detection is evaluated on two traffic scene event camera datasets using three standard metrics. Gen1 Dataset [50]: The Gen1 Automotive Detection Dataset comprises over 39 hours of 304 240 event video, captured across urban, highway, and rural traffic scenarios. It contains over 255,000 manual annotations for pedestrians and cars, annotated at a frequency of 1-4 Hz. 1Mpx Dataset [51]: The 1 Megapixel Automotive Detection Dataset provides over 14 hours of high-resolution event video and 25 million annotations for cars, pedestrians, and two-wheelers, making it ideal for developing advanced detection models in dynamic traffic scenes. Action Recognition is assessed on four representative datasets using three evaluation metrics. THUE-ACT-50-CHL Dataset [53]: The dataset is a very challenging dataset with 50 action classes and 2330 recordings... DVS Action Dataset [54]: The dataset contains 10 actions performed by 15 subjects... HMDB51-DVS & UCF101-DVS [55]: Event-based versions of HMDB51 [56] and UCF101 [57], converted using a DAVIS240 camera. |
| Dataset Splits | No | The paper describes the datasets used and their characteristics (e.g., number of events, classes, recordings) but does not specify the training, validation, or test splits for any of them. It refers to 'standard metrics' but not standard data partitioning. |
| Hardware Specification | Yes | The training pipeline is managed with Py Torch Lightning on NVIDIA GPUs... RT-DETR-R50 achieves 53.1% m AP while running at 108 FPS on an NVIDIA T4 GPU |
| Software Dependencies | No | The implementation relies on Py Torch [59] for the core framework and Py Torch Geometric (Py G) [60] for graph-based operations. The paper mentions these software packages but does not provide specific version numbers. |
| Experiment Setup | Yes | The model is optimized using the Adam W optimizer [61] with a One Cycle learning rate schedule [62]. Section D 'Hyperparameter Sensitivity Analysis' provides details on key hyperparameters: 'N=20,000 is selected as an effective balance point between accuracy and efficiency for the model', 'Adjusting nmin within a four-fold range (50 to 200) results in largely stable model performance', 'Our sensitivity analysis on this parameter [Ξ΄st] (see Table 6) reveals a key phenomenon of asymmetric sensitivity.' |