Robust and Decomposable Average Precision for Image Retrieval

Authors: Elias Ramzi, Nicolas THOME, Clément Rambour, Nicolas Audebert, Xavier Bitot

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments conducted on three image retrieval datasets show that ROADMAP outperforms several recent AP approximation methods and highlight the importance of our two contributions. Finally, using ROADMAP for training deep models yields very good performances, outperforming state-of-the-art results on the three datasets.
Researcher Affiliation Collaboration 1CEDRIC, Conservatoire National des Arts et Métiers, Paris, France 2Coexya, Paris, France
Pseudocode No The paper describes its methods using mathematical equations and textual explanations, but no pseudocode or algorithm blocks are provided.
Open Source Code Yes Code and instructions to reproduce our results will be made publicly available at https://github.com/elias-ramzi/ROADMAP.
Open Datasets Yes We evaluate ROADMAP on the following three image retrieval datasets: CUB-200-2011 [37]... Stanford Online Product (SOP) [31]... INaturalist-2018 [36].
Dataset Splits Yes We follow the standard protocol and use the first (resp. last) 100 classes for training (resp. evaluation). We use the reference train and test splits from [31]. We use the splits from [2] with 70% of the classes in the train set and the rest in the test set.
Hardware Specification No The paper mentions 'HPC resources of IDRIS under the allocation 2021AD011012645 made by GENCI' but does not provide specific hardware details like GPU/CPU models or memory specifications.
Software Dependencies No The paper mentions using 'Adam', 'AdamW', and 'PyTorch implementations' but does not specify version numbers for any of these software components.
Experiment Setup Yes For all experiments in Section 4.1 and Section 4.2, we use λ = 0.5 for LROADMAP in Eq. (2), τ = 0.01 and ρ = 100 for LSup AP in Eq. (5), α = 0.9 and β = 0.6 for Lcalibr. in Eq. (7). All models are trained in the same setting (Res Net-50 backbone, embedding size 512, batch size 64).