Datasets:
Tasks:
Visual Question Answering
Formats:
json
Size:
< 1K
Tags:
video-question-answering
spatio-temporal-reasoning
furniture-assembly
multiple-choice
cvpr-2026
License:
Update dataset card README
Browse files
README.md
CHANGED
|
@@ -1,3 +1,100 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
+
pretty_name: Flat-Pack Bench
|
| 4 |
+
task_categories:
|
| 5 |
+
- visual-question-answering
|
| 6 |
+
tags:
|
| 7 |
+
- video-question-answering
|
| 8 |
+
- spatio-temporal-reasoning
|
| 9 |
+
- furniture-assembly
|
| 10 |
+
- multiple-choice
|
| 11 |
+
- cvpr-2026
|
| 12 |
+
configs:
|
| 13 |
+
- config_name: default
|
| 14 |
+
default: true
|
| 15 |
+
data_files:
|
| 16 |
+
- split: test
|
| 17 |
+
path: questions/questions.jsonl
|
| 18 |
---
|
| 19 |
+
|
| 20 |
+
# Flat-Pack Bench 🧩
|
| 21 |
+
|
| 22 |
+
**Furniture assembly as a spatio-temporal stress test for large vision-language models.**
|
| 23 |
+
|
| 24 |
+
Flat-Pack Bench is a multiple-choice benchmark for evaluating fine-grained
|
| 25 |
+
spatio-temporal understanding in real furniture assembly videos. Each question
|
| 26 |
+
asks a model to reason about object parts, contact events, assembly order, final
|
| 27 |
+
connectivity, or part identity across time.
|
| 28 |
+
|
| 29 |
+
Project page: **https://flat-pack-bench.github.io**
|
| 30 |
+
|
| 31 |
+
## 🎯 Benchmark Tasks
|
| 32 |
+
|
| 33 |
+
The benchmark contains **602** multiple-choice questions organized into four
|
| 34 |
+
families:
|
| 35 |
+
|
| 36 |
+
| Family | Count | What it tests |
|
| 37 |
+
|---|---:|---|
|
| 38 |
+
| Temporal Localization | 103 | Identify which connection happened first, last, next, or most recently relative to an assembly state. |
|
| 39 |
+
| Temporal Ordering | 155 | Recover the order in which parts or part-pairs become physically connected. |
|
| 40 |
+
| Mating | 87 | Determine whether two labeled parts are connected in the fully assembled furniture. |
|
| 41 |
+
| Tracking | 257 | Match labeled parts across two states, requiring object identity through occlusion and motion. |
|
| 42 |
+
|
| 43 |
+
The Hugging Face Dataset Viewer is configured to display the benchmark question
|
| 44 |
+
table from `questions/questions.jsonl`. Other JSON files in this repository are
|
| 45 |
+
supporting annotations, mask files, and frame indexes rather than row-oriented
|
| 46 |
+
question examples.
|
| 47 |
+
|
| 48 |
+
## 📁 Repository Structure
|
| 49 |
+
|
| 50 |
+
| Path | Contents |
|
| 51 |
+
|---|---|
|
| 52 |
+
| `questions/` | Benchmark questions in JSONL form plus one YAML source file per question. |
|
| 53 |
+
| `videos/` | Raw, trimmed, and keyframe MP4 video variants. |
|
| 54 |
+
| `rgb-frames/` | Extracted RGB frames for videos, with source-frame-to-file mappings. |
|
| 55 |
+
| `segmentation-masks/` | Part-level masks encoded as compressed run-length encodings. |
|
| 56 |
+
| `furniture-annotations/` | Furniture part semantics, connectivity graphs, similar-part annotations, and OBJ part meshes. |
|
| 57 |
+
|
| 58 |
+
Most media and annotation paths follow this key pattern:
|
| 59 |
+
|
| 60 |
+
```text
|
| 61 |
+
<category>/<furniture_name>/<video_id>/
|
| 62 |
+
```
|
| 63 |
+
|
| 64 |
+
For example, the video ID `d9sENgGhBk4` for `Chair/ronninge` can be found under
|
| 65 |
+
matching paths in `videos/`, `rgb-frames/`, and, when available,
|
| 66 |
+
`segmentation-masks/`.
|
| 67 |
+
|
| 68 |
+
## 🗂️ Main Schemas
|
| 69 |
+
|
| 70 |
+
Detailed schema notes live next to the relevant data:
|
| 71 |
+
|
| 72 |
+
- `questions/README.md` documents `questions/questions.jsonl`.
|
| 73 |
+
- `questions/yamls/README.md` documents `questions/yamls/*.yaml`.
|
| 74 |
+
- `furniture-annotations/part-annotations/README.md` documents per-furniture part graphs and semantics.
|
| 75 |
+
- `rgb-frames/README.md` documents frame image paths and `frame_id_to_fn.json`.
|
| 76 |
+
- `segmentation-masks/README.md` documents mask JSON files.
|
| 77 |
+
- `videos/README.md` documents video variants and path conventions.
|
| 78 |
+
|
| 79 |
+
## 🙏 Acknowledgments
|
| 80 |
+
|
| 81 |
+
Flat-Pack Bench is built around real furniture assembly videos and annotations
|
| 82 |
+
from the IKEA Video Manuals ecosystem. Please also acknowledge the authors of
|
| 83 |
+
[IKEA Manuals at Work: 4D Grounding of Assembly Instructions on Internet Videos](https://yunongliu1.github.io/ikea-video-manual/):
|
| 84 |
+
Yunong Liu, Cristobal Eyzaguirre, Manling Li, Shubh Khanna, Juan Carlos
|
| 85 |
+
Niebles, Vineeth Ravi, Saumitra Mishra, Weiyu Liu, and Jiajun Wu.
|
| 86 |
+
|
| 87 |
+
## 📚 Citation
|
| 88 |
+
|
| 89 |
+
If you use Flat-Pack Bench, please cite:
|
| 90 |
+
|
| 91 |
+
```bibtex
|
| 92 |
+
@InProceedings{Chetan_2026_CVPR,
|
| 93 |
+
author = {Chetan, Aditya and Cai, Eric and Kushwaha, Peeyush and Kani, Bharath Raj Nagoor and Mall, Utkarsh and Wang, Qianqian and Snavely, Noah and Hariharan, Bharath},
|
| 94 |
+
title = {Flat-Pack Bench: Evaluating Spatio-Temporal Understanding in Large Vision-Language Models through Furniture Assembly},
|
| 95 |
+
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
|
| 96 |
+
month = {June},
|
| 97 |
+
year = {2026},
|
| 98 |
+
pages = {16624-16634}
|
| 99 |
+
}
|
| 100 |
+
```
|