Datasetpublic

NeqCene/ikea_asm

IKEA ASM (RGB mirror) This repository is a re-upload of the public RGB distribution of the IKEA ASM dataset. It preserves the category/view organization of the source archives and includes the official annotation, calibration, and split files needed to use the RGB videos. This is a dataset mirror, not a training or experiment artifact. It does not contain model checkpoints, extracted features, project code, or training configuration. The repository currently covers the RGB… See the full description on the dataset page: https://huggingface.co/datasets/NeqCene/ikea_asm.

sourceHugging Faceupdated 2mo agoView on Hugging Face

0likes1.7kdownloads

Dataset Card

IKEA ASM (RGB mirror)

This repository is a re-upload of the public RGB distribution of the IKEA ASM dataset. It preserves the category/view organization of the source archives and includes the official annotation, calibration, and split files needed to use the RGB videos.

This is a dataset mirror, not a training or experiment artifact. It does not contain model checkpoints, extracted features, project code, or training configuration. The repository currently covers the RGB streams and their official auxiliary files; it is not a claim to mirror every possible modality of the broader IKEA ASM release (for example, separate depth or pose packages).

Dataset Layout

There are four furniture categories and three RGB camera views:

Category	Views
`Kallax_Shelf_Drawer`	`dev1`, `dev2`, `dev3`
`Lack_Coffee_Table`	`dev1`, `dev2`, `dev3`
`Lack_Side_Table`	`dev1`, `dev2`, `dev3`
`Lack_TV_Bench`	`dev1`, `dev2`, `dev3`

Each RGB archive is an uncompressed tar file named ikea_asm_RGB_<Category>_<View>.tar. The archives expand to paths such as:

text

Lack_Side_Table/<scan>/dev3/images/scan_video.avi

The tar files contain the original RGB AVI streams. Select one view archive or download all three view archives for a category, depending on the application.

Repository Files

File	Purpose
`ikea_asm_RGB_<Category>_<View>.tar`	RGB videos for one category and view
`action_annotations.zip`	`gt_action.npy`, `gt_segments.json`, and `ikea_annotation_db_full`
`camera_params.zip`	RGB/depth camera calibration files
`utility_files.zip`	Action lists, official train/test split indexes, and calibration assets
`MANIFEST.tsv`	Archive category, view, file count, and payload size
`SHA256SUMS`	SHA-256 checksums for all 12 tar files and the three ZIP files

The three auxiliary archives are kept separate so that users who need only the videos do not have to download them.

Download

Install the Hugging Face CLI, then download one view archive:

bash

hf download NeqCene/ikea_asm \
  ikea_asm_RGB_Lack_Side_Table_dev3.tar \
  --repo-type dataset --local-dir .
tar -xf ikea_asm_RGB_Lack_Side_Table_dev3.tar

Download all RGB views for a category by listing the three archives explicitly:

bash

hf download NeqCene/ikea_asm \
  ikea_asm_RGB_Lack_Side_Table_dev1.tar \
  ikea_asm_RGB_Lack_Side_Table_dev2.tar \
  ikea_asm_RGB_Lack_Side_Table_dev3.tar \
  --repo-type dataset --local-dir .

Download the official auxiliary files when using annotations, camera calibration, or the provided splits:

bash

hf download NeqCene/ikea_asm \
  action_annotations.zip camera_params.zip utility_files.zip \
  --repo-type dataset --local-dir .
unzip action_annotations.zip
unzip camera_params.zip
unzip utility_files.zip

Verification

SHA256SUMS covers every archive in the repository. When only a subset has been downloaded, ignore entries for missing files:

bash

sha256sum -c SHA256SUMS --ignore-missing

For a complete mirror download, the same command verifies all 15 archives. MANIFEST.tsv records the number of files and the sum of their uncompressed payload bytes inside each tar archive; tar headers are not included in the payload total.

Notes on the Mirror

The Lack_Side_Table_dev3 tar was rebuilt from a verified copy of its original RGB AVI after an incomplete local copy was detected. Its archive structure, file count, payload size, and SHA-256 are recorded in MANIFEST.tsv and SHA256SUMS.

The archive format is intentionally preserved as tar rather than converting the videos to another container. Extraction is therefore expected before frame-level processing. The repository does not alter the annotation values or the official split definitions.

Source, Terms, and Citation

This mirror is based on the public IKEA ASM release. Follow the upstream dataset's terms of use and any applicable institutional or data-use requirements. Hosting this mirror does not grant additional rights or change the upstream license.

If you use IKEA ASM, cite:

text

Y. Ben-Shabat, X. Yu, F. Saleh, D. Campbell, C. Rodriguez-Opazo,
H. Li, and S. Gould, "The IKEA ASM Dataset: Understanding People
Assembling Furniture through Actions, Objects and Pose," WACV, 2021,
pp. 847-859.