# BUP-ST20 Dataset Structure

The following files relate to the BUP-ST20 dataset.
Below we briefly describe these files, how to uncompress them, and the general directory structure.

- 📦 Total download size: ~35 GB  
- 💾 Total uncompressed size: ~225 GB

## 1. RGB and Depth Images

These are split across multiple tar files and so to unzip it all together you
need to use the following command (on a Linux style environment). Please keep
in mind these are high resolution images and so will take up around 75Gb
unzipped.

For the RGB images:
```
cat bupst20_rgb.tar.gz.a* | tar -xzvf -
```

For the depth images:
```
cat bupst20_depth.tar.gz.a* | tar xzvf -
```

## 2. Annotations
The annotations are compressed into a single file, however, when they are
unzipped they will take up around 150Gb in total. This is because we store
results for instance-based semantic segmentation on each frame in a "raw"
format. To untar this file just run:

```
tar -xzvf bupst20_annotations.tar.gz
```

## 3. Odoms
The odoms are compressed into a single file, when
they are unzipped they will take up around 2Mb in total.
To untar this file just run:

```
tar -xzvf bupst20_odometry.tar.gz
```

## 4. Final directory structure
After unpacking all of the data you will have the following directory structure.

BUPST20/
  ├── annotations/
  ├── depth/
  ├── odoms/
  └── rgb/

Under each of the sub-directories (depth, rgb, annotations, odoms)
you will have a directory unique for each video sequence and within each
directory a file for each frame of the sequence, this is easiest to see for the
"rgb" images for sequence "100" (e.g. BUPST20/rgb/100). This is a short
sequence with just 39 frames. If you run ls you should get something like the
following.

```
ls -1 MOT_ECCV_2024/rgb/100/

1600936832187439.tiff
1600936832254172.tiff
.
.
.
.
.
```

## 5. Individual Files (frameid)
With reference to the above example, each file is named using its frameid,
which is filename without .tiff. This is is unique and matches across all of
the directories. That is frameid "1600936832187439" has a corresponding file in
"depth" and the other directories. When you produce outputs for a particular
frame you need to output it using the same frameid.

For the annotations sometimes there is not a file corresponding to a particular
frameid. When this is the case it is because there is no annotation for that
particular frameid.

## 6. Citation
If you use the dataset in your work we ask that you cite the following paper.

@Article{2025bupst20,
  author   = {Guclu, Esra and Halstead, Michael and Denman, Simon and McCool, Chris},
  title    = {Weakly Labelled Spatial-Temporal Sweet Pepper Data: enabling higher quality detection, segmentation, and tracking.},
  journal  = {The International Journal of Robotics Research},
  year     = {2025},
  volume   = {},
  pages    = {},
  month    = {},
  doi      = {10.1177/02783649251379093},
  publisher= {SAGE Publications Sage UK: London, England}
}