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..
Miss-ReID: Delivering Robust Multi-Modality Object Re-Identification Despite Missing Modalities
Authors: Xi ruida
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate our model s efficacy and superiority over state-of-the-art methods in various modality-missing scenarios, and our endeavors further propel multi-modality Re ID into real-world applications. |
| Researcher Affiliation | Academia | Ruida Xi State Key Laboratory of Electromechanical Integrated Manufacturing of High-Performance Electronic Equipment, Xidian University EMAIL |
| Pseudocode | Yes | Additionally, we summarize the overall training procedure in Algorithm 1 and illustrate the inference procedure in Fig. 4 under modality-missing conditions, which are available in Appendices A.1 and A.2. |
| Open Source Code | No | Our code will be released after the acceptance of our paper. |
| Open Datasets | Yes | To evaluate our method under modality-missing scenarios, we conducted comprehensive experiments on multi-modality object Re ID benchmarks (RGBNT201 [28] and RGBNT100 [40]) by introducing controlled data dropout during both training and inference phases. |
| Dataset Splits | Yes | To evaluate our method under modality-missing scenarios, we conducted comprehensive experiments on multi-modality object Re ID benchmarks (RGBNT201 [28] and RGBNT100 [40]) by introducing controlled data dropout during both training and inference phases. Consistent with conventions in Re ID community, two primary metrics Cumulative Matching Characteristics at Rank-1 (R-1 accuracy) and mean Average Precision(m AP) are employed to assess model performance under seven inference scenarios: one modality-complete scenario (denoted as RNT, where all data modalities RGB, Near Infrared and Thermal Infrared are fully available) and six modality-missing scenarios... |
| Hardware Specification | Yes | Our Miss-Re ID is implemented using Py Torch libraries and runs on a single NVIDIA RTX A6000 GPU with 48GB VRAM. |
| Software Dependencies | No | Our Miss-Re ID is implemented using Py Torch libraries and runs on a single NVIDIA RTX A6000 GPU with 48GB VRAM. In line with prior works [9, 10], the pre-trained CLIP [20] is applied for the vision and text encoders. |
| Experiment Setup | Yes | The model is trained in total of 50 epochs, with the L-d MMC module introduced after 20 epochs. We employ the Adam optimizer for training learnable modules, with a learning rate of 5e-3 for modality prompts and 3.5e-4 for others. The text encoder remains frozen throughout training. The number of structure-aware queries (Nq) is empirically set as 16 and 8 on RGBNT201 and RGBNT100, respectively. The length of modality prompts (Np) is experimentally set as 4 per modality. λ1 and λ2 in Eq. (11) are both experimentally set to 0.1. |