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..
FlexEvent: Towards Flexible Event-Frame Object Detection at Varying Operational Frequencies
Authors: Dongyue Lu, Lingdong Kong, Gim Hee Lee, Camille Simon Chane, Wei Ooi
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on large-scale event camera datasets demonstrate that our approach surpasses state-of-the-art methods, achieving significant improvements in both standard and high-frequency settings. Notably, our method maintains robust performance when scaling from 20 Hz to 90 Hz and delivers accurate detection up to 180 Hz, proving its effectiveness in extreme conditions. |
| Researcher Affiliation | Academia | 1National University of Singapore 2IPAL, CNRS IRL 2955, Singapore 2ETIS UMR 8051, CY Cergy Paris University, ENSEA, CNRS, France |
| Pseudocode | No | The paper describes its methodology in prose and mathematical formulas, but does not include any explicit sections or figures labeled 'Pseudocode' or 'Algorithm', nor any structured, code-like blocks. |
| Open Source Code | Yes | Project Page & Code: flexevent.github.io |
| Open Datasets | Yes | We conduct experiments based on three large-scale datasets: 1DSEC-Det [16], 2DSEC-Detection [50], and 3DSEC-MOD [66]. These datasets comprise 78,344 frames across 60 sequences, 52,727 frames over 41 sequences, and 13,314 frames within 16 sequences, respectively. making them suitable for evaluating event-based object detection methods. We prioritize DSEC-Det [16] as the primary benchmark for comparisons, as it is the largest, most recent, and most comprehensive eventframe perception dataset. |
| Dataset Splits | Yes | We conduct experiments based on three large-scale datasets: 1DSEC-Det [16], 2DSEC-Detection [50], and 3DSEC-MOD [66]. ... We prioritize DSEC-Det [16] as the primary benchmark for comparisons... We report results on Car and Pedestrian classes... For event-frame fusion methods, we compare DAGr [16] and HDI-Former [31] on DSEC-Det [16] using results from the original paper and retrain our method on DSEC-Detection [50] and DSEC-MOD [66] using standard settings to compare with CAFR [3] and RENet [66]. ... Each training sample contains a sequence length of 11 frames, allowing the model to learn temporal dependencies effectively. |
| Hardware Specification | Yes | Experiments are conducted on two NVIDIA RTX A5000 GPUs with 24GB memory, with the entire training process completed in approximately one day. ... All training experiments are carried out on two NVIDIA RTX A5000 GPUs, each with 24GB of memory... We compare inference times and parameter counts using an NVIDIA A5000 24GB GPU and an AMD EPYC 32-Core Processor. |
| Software Dependencies | No | Our training follows YOLO-X [63], using the standard detection loss consisting of Io U loss, classification loss, and regression loss, plus a lightweight regularizer that balances the contributions of each modality. ... The training process spans 100,000 iterations, utilizing the Adam optimizer [27] with a One Cycle learning rate schedule [44]... Consistent with RVT [18], we employ a mixed batching strategy... |
| Experiment Setup | Yes | Our training follows YOLO-X [63], using the standard detection loss consisting of Io U loss, classification loss, and regression loss, plus a lightweight regularizer that balances the contributions of each modality. The model is trained for 100,000 iterations with a batch size of 8 and a sequence length of 11, using a learning rate of 1e-4. Experiments are conducted on two NVIDIA RTX A5000 GPUs with 24GB memory, with the entire training process completed in approximately one day. |