AVAtt v2 release (part 3)

README.md

@@ -0,0 +1,120 @@
+---
+license: cc0-1.0
+pretty_name: AVAtt — Art Visual Attention
+language:
+  - en
+task_categories:
+  - image-to-image
+  - other
+tags:
+  - eye-tracking
+  - saliency
+  - saliency-prediction
+  - scanpath
+  - scanpath-prediction
+  - visual-attention
+  - gaze
+  - art
+  - paintings
+  - cultural-heritage
+size_categories:
+  - n<1K
+configs:
+  - config_name: default
+    data_files:
+      - split: train
+        path: data/train-*.parquet
+---
+
+# AVAtt — Art Visual Attention
+
+**Where do people look when they freely view paintings?** AVAtt provides individual eye-tracking scanpaths,
+fixation maps, saliency maps and temporal saliency videos for **283 paintings** from **16 art movements**, recorded
+from **152 observers**.
+
+| | |
+|---|---|
+| Paintings | 283 · 16 movement groups · 81 identified artists |
+| Scanpaths | 4,505 (10–32 observers per painting, median 15) |
+| Fixations | 121,426 |
+| Maps | binary fixation maps · saliency maps at σ = 1° (35.87 px) |
+| Temporal | 283 videos (60 fps, 5 s) of accumulated attention |
+| Metadata | title · artist · year · movement · genre · medium · copyright status |
+
+Code, benchmarks and the full analysis pipeline: **[github.com/kmamine/AVAtt---Art-Visual-Attention](https://github.com/kmamine/AVAtt---Art-Visual-Attention)**
+
+## Load
+
+```python
+from datasets import load_dataset
+ds = load_dataset("Amine-CV/AVAtt", split="train")
+row = ds[0]
+row["image"], row["fixation_map"], row["saliency_map"]   # PIL images, 1600×900
+row["scanpaths"]                                         # list of {observer, x, y, onset_ms, duration_ms, distance_mm}
+```
+
+Or with the companion package (also gives the videos, metrics and map utilities):
+
+```python
+pip install git+https://github.com/kmamine/AVAtt---Art-Visual-Attention.git
+import avatt
+ds = avatt.load()
+```
+
+## Files
+
+```
+metadata.csv                         one row per painting
+images/<id>.jpg                      1600×900 stimulus (painting centred on grey RGB 170)
+scanpaths/<id>.jsonl                 {"observer": k, "fixations": [{"x", "y", "onset_ms", "duration_ms", "distance_mm"}, ...]}
+fixation_maps/<id>.png               binary fixation map
+saliency_maps/<id>.png               Gaussian-smoothed fixation map, σ = 1° of visual angle
+temporal_saliency_maps/<id>.mp4      attention accumulated up to time t, 60 fps, 5 s
+data/train-*.parquet                 all of the above except the videos (powers the viewer and load_dataset)
+```
+
+Coordinates are canvas pixels. `painting_x0, y0, x1, y1` locate the painting inside the canvas.
+`onset_ms` is relative to the first fixation on the painting.
+
+## Collection
+
+- **Eye tracker**: Tobii X2-30 (30 Hz), 9-point calibration, Tobii I-VT fixation filter.
+- **Display**: 1920×1080, 531 mm wide.
+- **Viewing distance**: 56.8 ± 3.9 cm, recorded per fixation.
+- **Task**: free viewing, "freely view and enjoy the paintings".
+- **Presentation**: paintings shown for 5.5 s or 15 s, separated by a grey screen.
+- **Participants**: 152 volunteers.
+  - Age 23.5 ± 8.0 years, range 18–62.
+  - 13 countries of origin.
+  - 96% novices in art.
+  - Normal or corrected vision.
+- **Quality control**: scanpaths spanning ≥ 4.5 s and paintings with ≥ 10 valid scanpaths.
+
+## Metadata columns
+
+`image_id`, `title`, `artist`, `year`, `movement`, `movement_group`, `genre`, `medium`,
+`attribution_confidence` (high = identified work, medium = artist attributed, low = movement only),
+`copyright_status`, `exposure_s`, `n_observers`, `n_fixations`, `painting_x0/y0/x1/y1`, `duplicate_of`,
+`note`, `original_filename`.
+
+## License and images
+
+The gaze data, maps, videos and annotations are released under **CC0 1.0**. The painting images are not covered by
+this licence:
+- **248 works** are in the public domain.
+- **35 works** are in copyright or of unknown status; see `copyright_status`. They are included solely to support
+  non-commercial scientific research on visual attention.
+
+Rights holders can request removal by opening a discussion on this repository.
+
+## Citation
+
+```bibtex
+@inproceedings{kerkouri2024avatt,
+  title     = {{AVAtt}: Art Visual Attention dataset for diverse painting styles},
+  author    = {Kerkouri, Mohamed Amine and Tliba, Marouane and Chetouani, Aladine and Bruno, Alessandro},
+  booktitle = {Proceedings of the 2024 Symposium on Eye Tracking Research and Applications (ETRA '24)},
+  year      = {2024},
+  doi       = {10.1145/3649902.3655656}
+}
+```