Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval
Authors: Mohammad Omama, Po-han Li, Sandeep Chinchali
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted extensive experiments on four retrieval datasets, including Stanford Online Products (So P) and Pittsburgh30k, using four different off-the-shelf foundation models, including Dino V2 and CLIP. AE-SVC demonstrates up to a 16% improvement in retrieval performance, while (SS)2D shows a further 10% improvement for smaller embedding sizes. |
| Researcher Affiliation | Academia | Mohammad Omama, Po-han Li, Sandeep Chinchali The University of Texas at Austin EMAIL |
| Pseudocode | No | The paper describes the methodology for AE-SVC and (SS)2D in Section 3, detailing the steps and losses, but does not present this information in a structured pseudocode or algorithm block. |
| Open Source Code | No | The paper includes a reproducibility statement discussing theoretical analysis and hyperparameters but does not explicitly state that source code for the described methodology is being released or provide a link to a code repository. |
| Open Datasets | Yes | We evaluate our approach on four distinct image retrieval datasets: In Shop (Liu et al., 2016), Stanford Online Products (SOP) (Song et al., 2016), Pittsburgh30K (Torii et al., 2013), and Tokyo Val (Torii et al., 2015). In Shop and SOP are state-of-the-art (SOTA) image retrieval datasets commonly used in metric learning research, while Pittsburgh30K and Tokyo Val are recognized as SOTA place recognition datasets, frequently used in robotics research. |
| Dataset Splits | No | The paper mentions using a 'query set Q' and a 'reference set R' for image retrieval, and states that 'both AE-SVC and (SS)2D utilize only the reference set R during the training time. The query set Q is assumed to be unavailable to the user during training.' However, it does not specify the exact split percentages, sample counts, or the methodology used to create these splits from the underlying datasets. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models, CPU types, or memory used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific software dependency details, such as programming language versions or library versions (e.g., Python 3.x, PyTorch 1.x), that are needed to replicate the experiments. |
| Experiment Setup | Yes | We discussed four hyperparameters in Sec. 3.1: λrec, λcov, λvar, and λmean. We empirically determined the values of these hyperparameters, ensuring optimal performance across all datasets. The fixed values are: λrec = 25, λcov = 1, λvar = 15, and λmean = 1. |