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..
Neptune-X: Active X-to-Maritime Generation for Universal Maritime Object Detection
Authors: Yu Guo, Shengfeng He, Yuxu Lu, Haonan An, Yihang Tao, Huilin Zhu, Jingxian Liu, Yuguang Fang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that our approach sets a new benchmark in maritime scene synthesis, significantly improving detection accuracy, particularly in challenging and previously underrepresented settings. and 4 Experiments section. |
| Researcher Affiliation | Academia | Yu Guo1,3, Shengfeng He2, , Yuxu Lu4, Haonan An1, Yihang Tao1, Huilin Zhu5, Jingxian Liu3, Yuguang Fang1 1Hong Kong JC STEM Lab of Smart City and Department of Computer Science, City University of Hong Kong 2Singapore Management University 3State Key Laboratory of Maritime Technology and Safety, Wuhan University of Technology 4The Hong Kong Polytechnic University 5Wuhan University of Science and Technology Corresponding author: EMAIL. |
| Pseudocode | Yes | Algorithm 1 Bi OW-Attn and Algorithm 2 Data Sampling |
| Open Source Code | Yes | https://github.com/gy65896/Neptune-X and Once the blind review period is finished, we ll open-source all codes, instructions, and model checkpoints. |
| Open Datasets | Yes | To support training and evaluation under diverse maritime conditions, we construct a new benchmark, the Maritime Generation Dataset, which covers a broad range of scenarios with variations in object category, viewpoint, environment, and location. and Table 1: Data source of MGD. Source Imaging Viewpoint Num. Ma STr1325 [3] ship view 800 USVInland [6] ship view 1000 MIT Sea Grant [9] ship view 100 SMD [24] shore and ship view 400 Seaships [33] shore view 1500 Seagull [29] aerial view 2996 Fvessel [12] shore view 1500 La RS [52] shore, ship, and aerial view 1973 Others shore, ship, and aerial view 1631 MGD shore, ship, and aerial view 11900 and Justification: We use the publicly accessible dataset in Table 1. from the NeurIPS checklist. |
| Dataset Splits | Yes | Furthermore, MGD is split into training (7,140 samples), validation (2,380 samples), and test sets (2,380 samples) in a 3:1:1 ratio, with the validation and test sets combined for image generation evaluation. |
| Hardware Specification | Yes | The Neptune-X framework is implemented in Py Torch 1.13 (Python 3.8) and executed on a PC with 2 Intel(R) Xeon(R) Silver 4410Y CPUs and 4 NVIDIA 5880 Ada GPUs. |
| Software Dependencies | Yes | The Neptune-X framework is implemented in Py Torch 1.13 (Python 3.8) and executed on a PC with 2 Intel(R) Xeon(R) Silver 4410Y CPUs and 4 NVIDIA 5880 Ada GPUs. |
| Experiment Setup | Yes | In the training, we employ the Adam W optimizer with an initial learning rate of 5 10 5 for 100, 000 iterations (requiring 100 training hours), while we apply the standard data augmentation techniques, including random horizontal flipping and scale resizing. The patch size and batch size are 512 512 and 8 for model training. |