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..
Sparse-to-dense Multimodal Image Registration via Multi-Task Learning
Authors: Kaining Zhang, Jiayi Ma
ICML 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on MSCOCO, Google Earth, VIS-NIR and VIS-IR-drone datasets demonstrate that our method achieves remarkable performance on multimodal cases. |
| Researcher Affiliation | Academia | 1Electronic Information School, Wuhan University, Wuhan, China. Correspondence to: Jiayi Ma <EMAIL>. |
| Pseudocode | No | The paper describes algorithms using mathematical equations and text but does not provide structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is available at https: //github.com/KN-Zhang/SDME. |
| Open Datasets | Yes | We conduct experiments on four datasets, including MSCOCO (Lin et al., 2014), Google Earth (Zhao et al., 2021), VIS-NIR (Brown & S usstrunk, 2011) and VIS-IR-drone (Sun et al., 2022). |
| Dataset Splits | No | The paper specifies training and test splits for datasets (e.g., '7:3 ratio for training and test data generation') but does not explicitly mention a separate validation split or cross-validation setup. |
| Hardware Specification | Yes | All the experiments are conducted on a single TITAN RTX. |
| Software Dependencies | No | The paper mentions using 'Adam W' for training but does not provide specific version numbers for software libraries or frameworks like PyTorch, TensorFlow, or Python itself. |
| Experiment Setup | Yes | We train with Adam W with an initial learning rate of 1e 4, a weight decay of 5e 4, and a batch size of 8 image pairs. We train 100 epochs and use Cosine Annealing to schedule the learning rate. In the sparse branch, we calculate LAP by sampling positive samples within a radius of 3 pixels and negative ones between a radius of 5 and 7 pixels. The patch size is set to 16 to calculate Lcosim and Lpeaky. The total loss in this branch is set to LAP + 5Lcosim + Lpeaky + 0.008Lguide, while that in the dense branch is Lconv1 + Lconv2 + Lmc. |