Skip to content

Latest commit

 

History

History
100 lines (65 loc) · 2.21 KB

ReadMe.md

File metadata and controls

100 lines (65 loc) · 2.21 KB
Raw

Dataset

Download

  • NYUV2
  cd segmentation/src/datasets/nyuv2
  python prepare_dataset.py path_to_save_data
  
  for example: python prepare_dataset.py /home/data/shicaiwei/NYUV2
  • CityScapes
  cd segmentation/src/datasets/cityscapes 
  python prepare_dataset.py path_to_save_data path_to_save_processing_data

Build soft link

   cd segmentation
   mkdir data
   ln -s path_to_nyuv2_dataset ./data/NYUV2
   ln -s path_to_cityscape_datset ./data/CityScape

Inference

Download pretrain model

  • NYUV2: baiduyun itas

    • full folder contains the baseline model trained with complete multimodal data
    • mad_mar folder contains the models trained with mad and mar strategies

  • Cityscape

  • modify the path to pretrained model in 'test/test_missing.py' (Line 99 for NYUV2 and Line 109 for Cityscape)

  • run code

      cd test
  bash test_nyuv2.sh
  bash test_cityscape.sh

Training From Scratch

Train multimodal model

    bash train_nyuv2_full.sh 
    bash train_train_cityscape_full.sh

Train Get MMANet model for incomplete data model

    bash train_nyuv2_mmanet.sh 
    bash train_train_cityscape_mmanet.sh