Skip to content

Latest commit

 

History

History
31 lines (25 loc) · 2.73 KB

README.md

File metadata and controls

31 lines (25 loc) · 2.73 KB
Raw

RDNet for Semantic Segmentation (UperNet)

backbone/rdnet.py for detection and segmentation is the exact same file as rdnet.py for classification.

Tested Environment

  • python 3.9.18
  • torch 1.11.0
  • timm 0.9.6
  • mmcv-full 1.4.8
  • mmdet 2.25.3
  • mmcls 0.25.0
  • mmsegmentation 0.22.1

Model Zoo

Pretrained Models & logs (HFHub): https://huggingface.co/naver-clova-ocr/rdnet_upernet_ade20k_160k/tree/main

$Backbone$ $Params$ $FLOPs$ $mIOU^{SS}$ $mIOU^{MS}$ url
RDNet-T 81M 757G 47.6 48.6 ckpt, train_log, ms_eval_log
RDNet-S 50M 832G 48.7 49.8 ckpt, train_log, ms_eval_log
RDNet-B 87M 971G 49.6 50.5 ckpt, train_log, ms_eval_log

Train

torchrun --nproc_per_node=4 train.py configs/rdnet/rdnet_tiny.py --launcher pytorch

mutltiscale inference

sudo -H $(which torchrun) --nproc_per_node=4 test.py configs/rdnet/rdnet_tiny.py [ckpt-path] --aug-test --launcher pytorch --eval mIoU --tmpdir [your-tmpdir]