Repository files navigation

• python3
• pytorch==0.3.1
• opencv==3.3.1
• scipy, numpy, progress, protobuf
• joblib (for parallel processing data.)
• tqdm
• [optional] dominate

The code is written and tested inpython3, please install all requirements in python3.

• Download the training data.

  • Imgs, train annotaions, test annotations are all available atBaiduPan (passwd: wf18).
  • You can also download data from onedrive:onedrive(no passwd required).

• put training data indata/ and test annotation in put it inevaluation/wireframe/,

    unzip v1.1.zip    unzip pointlines.zip    unzip linemat.zip

• Data Structure
  Each .pkl file contains the annotated wireframe of an image, and it consists of the following variables:

  *.pkl|-- imagename: the name of the image|-- img:         the image data|-- points:      theset of pointsin the wireframe, each point is represented by its (x,y)-coordinatesin the image|-- lines:       theset of linesin the wireframe, each line is represented by the indices of its two end-points|-- pointlines:     theset of associated lines of each point|-- pointlines_index:       line indexes of linesin'pointlines'|-- junction:       the junction locations, derived from the'points' and'lines'|-- theta:      the angle values of branches of each junction

• visualizing the wireframe.
  After loading the .pkl file, you can run something like the following in Python to visualize the wireframe:

  foridx, (i,j)inenumerate(lines,start=0):x1,y1=points[i]x2,y2=points[j]cv2.line(im, (int(x1),int(y1)), (int(x2),int(y2)), (0,255,0),2,cv2.LINE_8)

• Preprocess data.

  cd juncpython3 main.py --create_dataset --exp 1 --jsoncd linepxpython3 main.py --genLine

Note:--json means you put the hype-parameters injunc/hypes/1.json.

• train junction detector.

  cd juncpython3 main.py --exp 1 --json --gpu 0 --balance

• train line pixel detecor.

  cd linepxpython3 main.py --netType stackedHGB --GPUs 0 --LR 0.001 --batchSize 4

• Test junction detector.

  cd juncpython3 main.py --exp 1 --json --test --checkepoch 16 --gpu 0 --balance

• Test line pixel detector.

  cd linepxpython3 main.py --netType stackedHGB --GPUs 0 --LR 0.001 --testOnly t

• Combine junction and line pixel prediction.

  python wireframe.py

The code for evaluation is put inevaluation/junc andevaluation/wireframe.Expected junction and wireframe precision/recall curve is like

For visualizing the result, we recommend generating an html file usingdominate tovisualize the result of different methods in columns.

@InProceedings{wireframe_cvpr18,author = {Kun Huang and Yifan Wang and Zihan Zhou and Tianjiao Ding and Shenghua Gao and Yi Ma},title = {Learning to Parse Wireframes in Images of Man-Made Environments},booktitle = {CVPR},month = {June},year = {2018}}

You can use this code/dataset for your research and other usages, following MIT License.