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 [1].
Bi-CMR: Bidirectional Reinforcement Guided Hashing for Effective Cross-Modal Retrieval
Authors: Tieying Li, Xiaochun Yang, Bin Wang, Chong Xi, Hanzhong Zheng, Xiangmin Zhou10275-10282
AAAI 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The Bi-CMR is evaluated by conducting extensive experiments over these two datasets. Experimental results prove the superiority of Bi-CMR over four state-of-the-art methods in terms of effectiveness. |
| Researcher Affiliation | Academia | Tieying Li1, Xiaochun Yang1*, Bin Wang1, Chong Xi1, Hanzhong Zheng1, Xiangmin Zhou2 1Northeastern University, China 2RMIT University, Australia EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes its framework and components, but it does not include any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | No | The paper does not provide any statement about releasing source code or a link to a code repository for the described methodology. |
| Open Datasets | Yes | We choose two commonly used datasets, MIRFlickr25K (Huiskes and Lew 2008) and NUS-WIDE10.5K... MIRFlickr25K2 is a benchmark dataset collected from Flickr. ... NUSWIDE10.5K is a dateset created by filtering NUS-WIDE 3. ... 2http://press.liacs.nl/mirflickr/mirdownload.html 3https://lms.comp.nus.edu.sg/wpcontent/uploads/2019/research/nuswide/NUS-WIDE.html |
| Dataset Splits | Yes | We construct two training sets by randomly selecting 10, 000 from MIRFlickr25K and 4, 000 from NUSWIDE10.5K. 2, 000 pairs are randomly selected as query set and the remaining as the retrieval database. ... To evaluate the four hyperparameters in Eq. 7, we construct validation set by randomly choosing 2, 000 data from database. |
| Hardware Specification | Yes | The training time for one epoch does not exceed 2 minutes on a single RTX2070 GPU. |
| Software Dependencies | No | The paper mentions PyTorch and ADAM optimizer but does not specify their version numbers or other software dependencies with versions. |
| Experiment Setup | Yes | The hidden sizes of the SSR network are set to 1, 024. ... The hash layer size for final hash codes is set as 16, 32, and 64 respectively. ... train our model using the ADAM optimizer with an initialized learning rate of 1e 4 in image, text network and 1e 3 in multilabel network with a batch size of 128, respectively. We set the maximum number of epochs as 120 to ensure the convergence. ... After every 20 epochs, the learning rate decreases by half. ... the best choice of α is around 0.01 0.05; β and γ is around 0.01, and their excessive disparity will lead to an imbalance in reinforcement; the optimal setting for δ is from 0.01 to 0.1. ... we set η to 0.9. ... the best choices for λ is around 0.4 on MIRFlickr25K and 0.7 on the NUS-WIDE10.5K. |