ExpandNets: Linear Over-parameterization to Train Compact Convolutional Networks

Authors: Shuxuan Guo, Jose M. Alvarez, Mathieu Salzmann

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we demonstrate the benefits of our Expand Nets on image classification, object detection, and semantic segmentation. We further provide an ablation study to analyze the influence of different expansion strategies and expansion rates in the supplementary material.
Researcher Affiliation Collaboration Shuxuan Guo CVLab, EPFL Lausanne 1015, Switzerland shuxuan.guo@epfl.ch Jose M. Alvarez NVIDIA Santa Clara, CA 95051, USA josea@nvidia.com Mathieu Salzmann CVLab, EPFL Lausanne 1015, Switzerland mathieu.salzmann@epfl.ch
Pseudocode No No pseudocode or algorithm blocks were found in the paper.
Open Source Code Yes Our code is available at https://github.com/GUOShuxuan/expandnets.
Open Datasets Yes For CIFAR-10 and CIFAR-100 [27], we use the same compact network as in [37]... For Image Net [46], we use the compact Mobile Net [21]... YOLO-LITE [23]... trained... on the PASCAL VOC2007 + 2012 [12, 13] training and validation sets... Cityscapes dataset [8].
Dataset Splits Yes trained the resulting network on the PASCAL VOC2007 + 2012 [12, 13] training and validation sets in the standard YOLO fashion [39, 40].
Hardware Specification Yes Epoch Time was evaluated on CIFAR-10 on 2 32G TITAN V100 GPUs.
Software Dependencies No No specific version numbers for software dependencies (e.g., PyTorch 1.9) were explicitly provided in the paper. It only mentions 'a pytorch implementation'.
Experiment Setup Yes We train for 350 epochs using a batch size of 128. We use stochastic gradient descent (SGD) with a momentum of 0.9, weight decay of 0.0005 and a learning rate of 0.1, divided by 10 at epochs 150 and 250.