Unleashing the Power of Contrastive Self-Supervised Visual Models via Contrast-Regularized Fine-Tuning

Authors: Yifan Zhang, Bryan Hooi, Dapeng Hu, Jian Liang, Jiashi Feng

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on image classification and semantic segmentation verify the effectiveness of Core-tuning.
Researcher Affiliation Collaboration Yifan Zhang1 Bryan Hooi1 Dapeng Hu1 Jian Liang2 Jiashi Feng3 1National University of Singapore 2Chinese Academy of Sciences 3SEA AI Lab
Pseudocode Yes The pseudo code is provided in the supplementary.
Open Source Code Yes The source code of Core-tuning is available at: https://github.com/Vanint/Core-tuning.
Open Datasets Yes Image Net20 (a subset of Image Net with 20 classes), CIFAR10, CIFAR100 [29], Caltech-101 [15], DTD [10], FGVC Aircraft [39], Standard Cars [28], Oxford-IIIT Pets [44] and Oxford 102 Flowers [42].
Dataset Splits Yes evaluated on val2012 set.
Hardware Specification No The paper mentions 'SGD based on two GPUs' but does not specify the type or model of GPUs or any other hardware components used for experiments.
Software Dependencies No The paper states 'We implement Core-tuning in Py Torch' but does not provide specific version numbers for PyTorch or other software dependencies.
Experiment Setup Yes Following [6], we perform parameter tuning for η and α from {0.1, 1, 10} on each dataset. Moreover, we set the temperature τ=0.07. To make the generated negative pairs closer to negatives, we clip λ Beta(α, α) by λ λn when generating hard negative pairs, where λn is a threshold and we set it to 0.8.