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..
MDReID: Modality-Decoupled Learning for Any-to-Any Multi-Modal Object Re-Identification
Authors: Yingying Feng, Jie Li, Jie Hu, Yukang Zhang, Lei Tan, Jiayi Ji
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments conducted on three challenging multi-modality Re ID benchmarks (RGBNT201, RGBNT100, MSVR310) consistently demonstrate the superiority of MDRe ID. Notably, MDRe ID achieves significant m AP improvements of 9.8%, 3.0%, and 11.5% in general modality-matched scenarios, and average gains of 3.4%, 11.8%, and 10.9% in modality-mismatched scenarios, respectively. |
| Researcher Affiliation | Academia | 1Northeastern University 2Xiamen University 3National University of Singapore EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methodology using textual explanations and mathematical formulations, along with a high-level framework diagram in Figure 2. However, it does not include any explicitly labeled 'Pseudocode' or 'Algorithm' block with structured steps. |
| Open Source Code | Yes | The code is available at: https://github.com/stone96123/MDRe ID. |
| Open Datasets | Yes | We evaluate the proposed MDRe ID framework on three publicly available datasets spanning both person (RGBNT201 [24]) and vehicle re-identification (RGBNT100 [8] and MSVR310 [26]) tasks. |
| Dataset Splits | Yes | RGBNT201 [24]: The dataset is split into 141 identities for training, 30 for validation, and 30 for testing, with the entire test set serving as the gallery and 10 records per identity sampled as probes. RGBNT100 [8]: Fifty vehicles (8,675 triples) are used for training, and the remaining 50 vehicles (8,575 triples) form the testing/gallery set, with 1,715 triples randomly selected as queries. MSVR310 [26]: The dataset is divided into 1,032 samples from 155 vehicles for training and 1,055 samples from 155 vehicles for testing/gallery, with 591 samples (from 52 vehicles) used as queries. |
| Hardware Specification | Yes | All experiments are implemented in Py Torch and conducted on a single NVIDIA RTX 4090 GPU with CUDA 12.5 and Python 3.8. |
| Software Dependencies | Yes | All experiments are implemented in Py Torch and conducted on a single NVIDIA RTX 4090 GPU with CUDA 12.5 and Python 3.8. |
| Experiment Setup | Yes | Input images are resized to 256 128 for RGBNT201 and 128 256 for RGBNT100 and MSVR310. We employ random horizontal flipping, cropping, and erasing [38] for data augmentation. The model is optimized using Adam with a batch size of 64. The base learning rate is initialized at 3.5 10 4, while the visual encoder is fine-tuned with a reduced rate of 5 10 6. Training is performed for 50 epochs. |