Contrastive Fine-grained Class Clustering via Generative Adversarial Networks

Authors: Yunji Kim, Jung-Woo Ha

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 4 EXPERIMENTS
Researcher Affiliation Industry Yunji Kim NAVER AI Lab yunji.kim@navercorp.com; Jung-Woo Ha NAVER AI Lab & NAVER CLOVA jungwoo.ha@navercorp.com
Pseudocode No The paper describes the model architecture and objective functions using equations and tables, but it does not include a block explicitly labeled as 'Pseudocode' or 'Algorithm'.
Open Source Code Yes Code is available at https://github.com/naver-ai/c3-gan.
Open Datasets Yes We tested our method on 4 datasets that consist of single object images. i) CUB (Wah et al., 2011): 5,994 training and 5,794 test images of 200 bird species. ii) Stanford Cars (Krause et al., 2013): 8,144 training and 8,041 test images of 196 car models. iii) Stanford Dogs (Khosla et al., 2011): 12,000 training and 8,580 test images of 120 dog species. iv) Oxford Flower (Nilsback & Zisserman, 2008): 2,040 training and 6,149 test images of 102 flower categories.
Dataset Splits No For example, for CUB: '5,994 training and 5,794 test images'. The paper lists training and test image counts for each dataset but does not explicitly mention a separate validation split or how it's handled for training.
Hardware Specification Yes The training was done with 2 NVIDIA-V100 GPUs
Software Dependencies No The paper mentions 'Adam optimizer' and 'Inception networks' but does not specify any software names with version numbers for libraries or programming languages used.
Experiment Setup Yes The weights of the loss terms (λ0, λ1, λ2, λ3, λ4) are set as (5, 1, 1, 0.1, 1), and the temperature τ is set as 0.1. We utilized Adam optimizer of which learning rate is 0.0002 and values of momentum coefficients are (0.5, 0.999).