Skip to content

Latest commit

 

History

History
43 lines (40 loc) · 1.14 KB

DATA.md

File metadata and controls

43 lines (40 loc) · 1.14 KB
Raw

Data Format

1. Input RGB-D Sequence

In SgSlam and ScanNet, their data sequence are collected from RGB-D sensors. The data format is as follow,

|---ab0201_03a
    |---color
    |---depth
    |---pose
    |---intrinsic
    |---prediction_no_augment
        |---frame-000000_label.json # Image taging and detection results
        |---frame-000000_mask.png # Pixel-wise instance mask
        |--- ...

In the sequence name ab0201_03a, ab0201 is a scene id and 03a is a scanning index.

2. Output Results

At the end of each sequence, the mapping node save the results in a pre-defined output folder. The output format is as below.

|---output
    |---ab0201_03a
        |---instance_map.ply # points are colored instance-wise
        |---instance_info.txt
        |---instance_box.txt
        |---${idx}.ply (intanace-wise point cloud)
        |--- ...

3. Dataset Folder

We organize the dataset folder as below.

${SGSLAM_DATAROOT}
|---splits
    |---val.txt
|---scans
    |---ab0201_03a
    |--- ...
|---output
    |---online_mapping # mapping results

ScanNet dataset folder is organized in the same way.