Home

Awesome

Pytorch implementation of UniNet (ECCV 2022)

tenser performance

This repo is the offcial implementation of the paper UniNet: Unified Architecture Search with Convolution, Transformer, and MLP

@article{UniNet,
  author  = {Jihao Liu, Xin Huang, Guanglu Song, Yu Liu, Hongsheng Li},
  journal = {arXiv:2207.05420},
  title   = {UniNet: Unified Architecture Search with Convolution, Transformer, and MLP},
  year    = {2022},
}

Update

20/12/2022 Update pretrained models.

25/10/2022 Update the source code.

Environment

The code is tested with torch==1.11 and timm==0.5.4.

Availble models

ModelsParams (M)FLOPs (G)Pretrain EpochsTop-1 Acc.ckpt
UniNet-B111.51.130081.0ckpt

Run experiments

Currently, we supporting running experiments with slurm. You can reproduce the results of UniNet-B1 as follows:

sh exp/b1/run.sh partition 8