search

SamsungLabs / FACaP

[IROS 2022]Floorplan-Aware Camera Poses Refinement
MIT License
21 stars 6 forks source link

To Organize Scans Data #3

Open hyunJIN7 opened 1 year ago

hyunJIN7 commented 1 year ago

Hi! Thanks for great work! I wonder how to organized the training data (floorplan.npy , segmentation). I have a few questions in this regard.

  1. Dose floorplan.npy consist of actual scale? I think find_nearest_segment is considering the acutal wall_pcd scale, it that right? Can you tell me how to make a custom floorplan.npy?

  2. Can I get frame-wall.png and frame-floor.png , the reseult of segmantation, by applying RGB images to MSeg as mentioned in the paper?

Thank you.

merlushka commented 1 year ago

@hyunJIN7 Hi! Sorry for late response.

  1. Yes, floorplan.npy is a plan in actual scale. It should contain the 2D coordinates of the edges of the scene plan (Nx4 array, each row is (p_1, p_2), where p_1, p_2 are the @d coordinates of two vertices (corners) connected by the edge (wall)
  2. Yes, frame-wall.png and frame-floor.png may be obtained by applyong any segmentation algorithm, for example, MSeg.