Discrete Image Hashing Using Large Weakly Annotated Photo Collections

Authors: Hanwang Zhang, Na Zhao, Xindi Shang, Huanbo Luan, Tat-seng Chua

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through training on one million weakly annotated images, our experimental results demonstrate that image retrieval using the proposed hashing method outperforms the other state-of-the-art ones on image and video benchmarks.
Researcher Affiliation Academia National University of Singapore Tsinghua University
Pseudocode Yes Algorithm 1: Discrete Weakly-Supervised Hashing
Open Source Code No The paper states that comparison methods were implemented using codes provided by their authors, but does not state that the authors of this paper are releasing their own code for the described methodology.
Open Datasets Yes For training the hashing function, we used a large weakly-annotated image dataset called SBU (Ordonez, Kulkarni, and Berg 2011).
Dataset Splits No The paper mentions a train/test split but does not explicitly detail a validation split or its size/methodology.
Hardware Specification Yes On our Intel i7 6-core machine with 3.0Ghz CPU and 64-GB memory, we needed about 1 hour per iteration for training 1M data.
Software Dependencies No The paper mentions "De CAF deep learning visual features for images (Donahue et al. 2013)" but does not provide specific version numbers for this or any other software component used in the experiments.
Experiment Setup Yes We empirically set λ2 to 1e 4... we set λ1 to the value with best performance. For the learning rate η used in U-subproblem, we initially set it to 1e 3 and used a dynamic learning rate updating heuristic (Qian 1999).