shicaiwei123/TMM-MMH

Code for Privileged Modality Learning via Multimodal Hallucination (TMM)

TMM-MMH

Code for Privileged Modality Learning via Multimodal Hallucination (TMM)

Face anti-spoofing task

DATASET

• DOWNLODAD LINK https://sites.google.com/view/face-anti-spoofing-challenge/welcome/challengecvpr2019
  • Require sign a agreement

Run

• parameter

  • data_root: path to your dataset, such as '/home/bbb/shicaiwei/data/liveness_data/CASIA-SURF'
  • modal: the modality used, such as rgb,depth,ir
  • backbone: name of backbone, it is only the name determined by your backbone, and can not to select backbone

• single baseline

  cd src
  bash baseline_depth.sh
  bash baseline_ir.sh
  bash baseline_rgb.sh

• multimodal teacher model

  cd src
  bash baseline_multi.sh
  bash cefa_baseline_multi.sh

• multimodal model hallucination

  • RAD

    cd src
    bash surf_patch_spp.sh
    bash cefa_patch_kd_spp.sh

  • DAD

    cd src
    bash surf_patch_feature.sh
    bash cefa_patch_kd_feature.sh

Action Recognition

Dataset

• NW-UCLA
• NTUD60
• UCF101
• HMDB51

RUN

• multimodal teacher model
  • see code in https://github.com/shicaiwei123/pytorch-i3d
• multimodal model hallucination
  • UCF101

    • cd ucf101/src
      bash ucf101_patch_spp.sh

  • HMDB51

    • cd hmdb51/src
      bash hmdb51_patch_spp.sh

  • NWUCLA

    • cd nw_ucla_cv
      bash ucla_patch_spp.sh

  • NTUD60

    • cd ntud60_cs
      bash ntud60_patch_spp.sh