docs: refresh IMPACT v1.1 dataset card
Browse files
README.md
CHANGED
|
@@ -3,57 +3,88 @@ license: cc-by-nc-sa-4.0
|
|
| 3 |
pretty_name: IMPACT
|
| 4 |
task_categories:
|
| 5 |
- video-classification
|
|
|
|
| 6 |
- other
|
| 7 |
tags:
|
| 8 |
- industrial assembly
|
| 9 |
- procedural action understanding
|
| 10 |
- temporal action segmentation
|
| 11 |
-
-
|
|
|
|
| 12 |
- RGB-D
|
| 13 |
- eye tracking
|
| 14 |
---
|
| 15 |
|
| 16 |
# IMPACT v1.1
|
| 17 |
|
| 18 |
-
IMPACT is a synchronized five-view RGB-D dataset and benchmark for multi-granularity human procedural action understanding in industrial assembly.
|
| 19 |
|
| 20 |
- [Project page](https://kratos-wen.github.io/IMPACT/)
|
| 21 |
-
- [Benchmark code](https://github.com/Kratos-Wen/IMPACT)
|
| 22 |
- [Google Drive mirror](https://drive.google.com/drive/folders/1P7vBnxSVH9g_lQc5n0c0WA47QAGEhE_U?usp=sharing)
|
| 23 |
|
| 24 |
-
##
|
| 25 |
|
| 26 |
-
**July 2026
|
| 27 |
|
| 28 |
-
## Tasks
|
| 29 |
|
| 30 |
-
|
| 31 |
-
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
- Procedure Step Recognition (`PSR`)
|
| 37 |
-
- Assembly State Recognition (`ASR`)
|
| 38 |
-
- Procedural Phase Recognition for the Left and Right Hands (`PPR-L/R`)
|
| 39 |
-
- Anomaly Type Recognition for the Left and Right Hands (`ATR-L/R`)
|
| 40 |
|
| 41 |
-
|
| 42 |
|
| 43 |
-
|
| 44 |
|
| 45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
|
| 47 |
```bash
|
| 48 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
```
|
| 50 |
|
| 51 |
-
The
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
|
| 53 |
-
##
|
| 54 |
|
| 55 |
-
|
| 56 |
|
| 57 |
## License
|
| 58 |
|
| 59 |
-
Dataset assets and documentation are licensed under [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/).
|
|
|
|
| 3 |
pretty_name: IMPACT
|
| 4 |
task_categories:
|
| 5 |
- video-classification
|
| 6 |
+
- feature-extraction
|
| 7 |
- other
|
| 8 |
tags:
|
| 9 |
- industrial assembly
|
| 10 |
- procedural action understanding
|
| 11 |
- temporal action segmentation
|
| 12 |
+
- cross-view understanding
|
| 13 |
+
- action forecasting
|
| 14 |
- RGB-D
|
| 15 |
- eye tracking
|
| 16 |
---
|
| 17 |
|
| 18 |
# IMPACT v1.1
|
| 19 |
|
| 20 |
+
IMPACT is a synchronized five-view RGB-D dataset and benchmark for multi-granularity human procedural action understanding in industrial assembly. It contains **112 trials from 13 participants** and **39.5 video hours** across one egocentric and four exocentric views.
|
| 21 |
|
| 22 |
- [Project page](https://kratos-wen.github.io/IMPACT/)
|
| 23 |
+
- [Benchmark code and task protocols](https://github.com/Kratos-Wen/IMPACT)
|
| 24 |
- [Google Drive mirror](https://drive.google.com/drive/folders/1P7vBnxSVH9g_lQc5n0c0WA47QAGEhE_U?usp=sharing)
|
| 25 |
|
| 26 |
+
## Release Update
|
| 27 |
|
| 28 |
+
**July 2026, v1.1.** All 560 TAS-B annotation files were revalidated, and 117 of 560 RGB videos (20.9%) were updated. Dependent task assets and selected TAS-S boundaries were regenerated from the refined canonical annotations. See [`v1.1/CHANGELOG.md`](v1.1/CHANGELOG.md) for the complete record.
|
| 29 |
|
| 30 |
+
## Benchmark Tasks
|
| 31 |
|
| 32 |
+
| Family | Tasks | Metrics |
|
| 33 |
+
| --- | --- | --- |
|
| 34 |
+
| Temporal Understanding | `TAS-S`, `TAS-BL/BR` | Accuracy, Edit, F1@{10,25,50} |
|
| 35 |
+
| Cross-View Understanding | `CV-TA`, `CV-SMR`, `CV-SMC` | Recall@{1,5}, Median Rank, Coverage, mAP, Top-1, Macro-F1 |
|
| 36 |
+
| Action Forecasting | `AF-S`, `AF-L` | mR@5, AUED, ED@z |
|
| 37 |
+
| State & Reasoning | `PSR`, `ASR`, `PPR-L/R`, `ATR-L/R` | task-specific F1, delay, POS, state, phase, and multi-label metrics |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
|
| 39 |
+
Formal task definitions, split semantics, baselines, and evaluation code are versioned in the [benchmark repository](https://github.com/Kratos-Wen/IMPACT/blob/main/docs/BENCHMARK.md).
|
| 40 |
|
| 41 |
+
## Download
|
| 42 |
|
| 43 |
+
The `v1.1/` directory provides independent archives for annotations, RGB videos, depth streams, ego audio, eye tracking, and I3D, MViTv2, and VideoMAEv2 features. Every ZIP extracts into the same `IMPACT-v1.1/` root.
|
| 44 |
+
|
| 45 |
+
The 2.8 GiB quick-start sample is available at `v1.1/sample/IMPACT-v1.1-sample.zip`. It contains three complete executions with synchronized media, all released feature types, and corresponding task annotations.
|
| 46 |
+
|
| 47 |
+
Selective download and automatic checksum verification:
|
| 48 |
|
| 49 |
```bash
|
| 50 |
+
git clone https://github.com/Kratos-Wen/IMPACT.git
|
| 51 |
+
cd IMPACT
|
| 52 |
+
python -m pip install -r requirements-tools.txt
|
| 53 |
+
python tools/download_impact.py --include sample --output downloads
|
| 54 |
+
python tools/verify_release.py downloads/v1.1/sample/IMPACT-v1.1-sample.zip
|
| 55 |
+
```
|
| 56 |
+
|
| 57 |
+
Use `python tools/download_impact.py --list` to inspect all bundle keys. Exact byte sizes and SHA-256 checksums are published in `v1.1/MANIFEST.tsv` and `v1.1/SHA256SUMS`.
|
| 58 |
+
|
| 59 |
+
## Data Structure
|
| 60 |
+
|
| 61 |
+
```text
|
| 62 |
+
IMPACT-v1.1/
|
| 63 |
+
├── annotations/{TAS-S,TAS-B,AF-S,ASR,PSR,PPR,ATR}/
|
| 64 |
+
├── videos/{ego,front,left,right,top}/
|
| 65 |
+
├── depth/{front,left,right,top}/
|
| 66 |
+
├── audio/ego/
|
| 67 |
+
├── eye_tracking/ego/
|
| 68 |
+
├── features/{I3D,MViTv2,VideoMAEv2}/
|
| 69 |
+
└── metadata/
|
| 70 |
```
|
| 71 |
|
| 72 |
+
The canonical identifier is `<execution_id>_<view>`. All views from the same trial are co-assigned in official splits to prevent cross-view leakage.
|
| 73 |
+
|
| 74 |
+
## Features
|
| 75 |
+
|
| 76 |
+
Released arrays contain frozen I3D (1024-D), MViTv2-B (768-D), and VideoMAEv2-G (1408-D) representations. Extraction conventions and reference scripts are documented in the [feature guide](https://github.com/Kratos-Wen/IMPACT/blob/main/docs/FEATURES.md).
|
| 77 |
+
|
| 78 |
+
## Ethics and Intended Use
|
| 79 |
+
|
| 80 |
+
Participants provided informed consent for data collection and research release. Public identifiers are anonymized, and the release excludes direct personal identifiers. The dataset is intended for non-commercial research and education in video understanding, procedural reasoning, forecasting, and industrial human-assistance systems. It must not be used to identify participants or infer sensitive personal attributes.
|
| 81 |
+
|
| 82 |
+
IMPACT represents a controlled industrial assembly setting and should not be treated as evidence that a model is safe for autonomous deployment. Users should assess domain shift, demographic coverage, failure modes, and human oversight requirements in their target environment. See the repository's [Ethics and Intended Use](https://github.com/Kratos-Wen/IMPACT/blob/main/docs/ETHICS_AND_USE.md) statement.
|
| 83 |
|
| 84 |
+
## Citation
|
| 85 |
|
| 86 |
+
The manuscript is under review at the ACM Multimedia 2026 Dataset Track. Stable versioned author metadata is available in [`CITATION.cff`](https://github.com/Kratos-Wen/IMPACT/blob/main/CITATION.cff); proceedings metadata will be added after publication.
|
| 87 |
|
| 88 |
## License
|
| 89 |
|
| 90 |
+
Dataset assets and release documentation are licensed under [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/). Benchmark code is distributed separately under Apache-2.0, and third-party methods retain their respective upstream terms.
|