Aligning Geometric Spatial Layout in Cross-View Geo-Localization via Feature Recombination

Authors: Qingwang Zhang, Yingying Zhu

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show that our proposed FRGeo not only achieves state-of-the-art performance on cross-view geo-localization benchmarks, including CVUSA, CVACT, and VIGOR, but also is significantly superior or competitive in terms of computational complexity and trainable parameters.
Researcher Affiliation Academia College of Computer Science and Software Engineering, Shenzhen University, China zhangqingwang2022@email.szu.edu.cn, zhuyy@szu.edu.cn
Pseudocode No The paper describes the Feature Recombination Module (FRM) and Weighted (B + 1)-tuple loss (WBL) using mathematical formulations and descriptive text, but no pseudocode or algorithm blocks are provided.
Open Source Code No Our project homepage is at https://zqwlearning.github.io/FRGeo.
Open Datasets Yes We evaluate our method on three public cross-view geo-localization datasets: CVUSA (Zhai et al. 2017), CVACT (Liu and Li 2019) and VIGOR (Zhu, Yang, and Chen 2021b).
Dataset Splits Yes CVUSA contains 35,532 image pairs for training and 8,884 image pairs for testing. This dataset consists of images mainly collected at suburban areas. CVACT provides 35,532 image pairs for training and 8,884 image pairs for validation (CVACT val). It also provides 92,802 image pairs to support fine-grained city-scale geo-localization (CVACT test).
Hardware Specification Yes We train the model on a NVIDIA V100 Server with Adam W (Loshchilov and Hutter 2017) optimizer.
Software Dependencies No The paper mentions using Conv Ne Xt-T as the backbone and AdamW as the optimizer, but it does not specify software versions for programming languages, libraries (e.g., PyTorch, TensorFlow), or CUDA.
Experiment Setup Yes α is set 10 in Equation (12). We train the model on a NVIDIA V100 Server with Adam W (Loshchilov and Hutter 2017) optimizer. To enable a fair comparison, the choice of hyperparameters and training strategy for all subsequent models remain entirely consistent with those of the Baseline.