Skip to content

Latest commit

 

History

History
90 lines (69 loc) · 3.3 KB

README_eft_vis.md

File metadata and controls

90 lines (69 loc) · 3.3 KB
Raw

Getting Started with EFT Dataset

Visualizing EFT Fitting Results

  • Run "demo/visEFTFit.py"
  • set EFT fitting dir path and image path.
python -m demo.visEFTFit --rendermode geo --img_dir (your_img_dir) --fit_data (downloaded_fitting_location)
#for example
python -m demo.visEFTFit --rendermode geo --img_dir ~/data/coco/train2014 --fit_data eft_fit/COCO2014-Part-ver01.json

  • The rendered ouptut is written in default location "render_eft". You can change it with "--render_dir (outputFolderName)"

  • More examples with other options:

#Render via Phong shading, on original image space
python -m demo.visEFTFit --rendermode geo

#Render via denspose IUV, on bbox image
python -m demo.visEFTFit --rendermode denspose --onbbox

#Use --waitforkeys to stop after visualizing each sample, waiting for any key pressed
python -m demo.visEFTFit --rendermode geo --waitforkeys

#use --turntable to show the mesh via turn table views
python -m demo.visEFTFit --rendermode geo --turntable

#Use --onbbox, if you want to visualize output in bbox space 
python -m demo.visEFTFit --onbbox

#Use --multi to show all annotated humans (multiple people) per image
python -m demo.visEFTFit --rendermode normal --multi

#Use --multi --turntable
python -m demo.visEFTFit --rendermode geo --multi --turntable

  • By default, the rendered images are saved in the directory (default: ./render_eft ) specified by "--render_dir"

  • Our visualization pipeline uses OpenGL, requiring a screen to dispaly output. For screenless rendering (e.g., a server without a screen), use "xvfb-run"

xvfb-run python -m demo.visEFTFit
  • In the EFT files for COCO, you can find coco annotation ID in 'annotId'. Check the following example script to visualize EFT fitting with other COCO annotations (e.g., bbox, skeleton)
python -m demo.visEFTFit_coco --cocoAnnotFile (your_coco_path)/annotations/person_keypoints_train2014.json

Visualizing EFT Fitting with GUI mode

  • In this visualization, you can use your keyboard and mouse to see the fitting output from any view point.

  • Script: "demo/visEFTFit_gui.py"

  • Run,

 python -m demo.visEFTFit_gui --img_dir /your/dataset/path --fit_dir /your/dataset/path 

 #For example
 python -m demo.visEFTFit_gui --img_dir ~/data/coco/train2014 --fit_dir ~/CVPR2020_submit_fits/11-08_coco_with8143
  • The default visualization will show a single person at each time
  • You can use "--multi" option to visualize all (reconstructed) humans for an image
  python -m demo.visEFTFit_gui --multi --img_dir ~/data/coco/train2014 --fit_dir ~/CVPR2020_submit_fits/11-08_coco_with8143
  • If you can see a 3D view, good! You can use mouse and keyboard to change viewpoint
  • In the 3d view, press 'q' to go to the next sample
  • Other key information:
    • mouse left + move: view change
    • mouse right + move: zoom in/out
    • shift + mouse left + move: pan
    • 'C': toggle between 3D view and image view
    • 'q': go to the next sample
    • 'w': toggle between solid mesh and wire-frame mesh
    • 'j': on/off for 3d skeleton
    • 'm': on/off for 3d mesh
    • 'f': on/off for floor

License

CC-BY-NC 4.0. See the LICENSE file.