Upload 5 files

.gitattributes

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+records.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,96 @@
+---
+pretty_name: ReGuLaR
+task_categories:
+- visual-question-answering
+language:
+- en
+size_categories:
+- 100K<n<1M
+---
+
+# ReGuLaR Data
+
+This repository contains the ReGuLaR training mix with images, question-answer pairs,
+bounding boxes, and relation annotations. Metadata is provided as parquet shards; images
+are provided as compressed source-specific archives to keep the repository uploadable.
+
+## Files
+
+```text
+data/train-xxxxx-of-yyyyy.parquet
+image_archives/*.tar.zst
+records.jsonl
+manifest.json
+stats.json
+extract_images.sh
+```
+
+The parquet schema contains:
+
+- `id`, `source`, `dataset`
+- `image`: primary image path, relative to the dataset root after extraction
+- `images`: all image paths for the sample; mGrounding samples may contain multiple images
+- `question`, `answer`
+- `relations_json`: subject-relation-object annotations serialized as JSON
+- `scene_graph_json`: compact scene graph or relation evidence serialized as JSON
+- `metadata_json`: source-specific metadata serialized as JSON
+
+## Download And Extract Images
+
+```bash
+from huggingface_hub import snapshot_download
+snapshot_download("zihuwang/ReGuLaR", repo_type="dataset", local_dir="regular_data")
+```
+
+Then extract the image archives from the dataset root:
+
+```bash
+cd regular_data
+bash extract_images.sh
+```
+
+Equivalently, you can extract archives manually:
+
+```bash
+for f in image_archives/*.tar.zst; do tar --zstd -xf "$f"; done
+```
+
+This creates the `images/` directory expected by the metadata paths:
+
+```text
+images/gqa/...
+images/openimage/...
+images/clevr/...
+images/visual_genome/...
+images/psg/...
+images/mgrounding/...
+images/vsr/...
+```
+
+## Usage
+
+```python
+import json
+from pathlib import Path
+from PIL import Image
+from datasets import load_dataset
+
+root = Path("regular_data")
+ds = load_dataset("parquet", data_files=str(root / "data/*.parquet"), split="train")
+
+ex = ds[0]
+image = Image.open(root / ex["image"]).convert("RGB")
+all_images = [Image.open(root / p).convert("RGB") for p in ex["images"]]
+
+relations = json.loads(ex["relations_json"])
+scene_graph = json.loads(ex["scene_graph_json"])
+metadata = json.loads(ex["metadata_json"])
+```
+
+Each relation follows a subject-relation-object format. Subject and object entries
+include names and bounding boxes when available. Bounding boxes use `bbox_xyxy`.
+
+## Notes
+
+The dataset combines examples derived from GQA, OpenImage, CLEVR, Visual Genome, PSG,
+mGrounding, and VSR. Please follow the licenses and terms of the original data sources.

extract_images.sh

@@ -0,0 +1,9 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+cd "$(dirname "$0")"
+for archive in image_archives/*.tar.zst; do
+  [ -e "$archive" ] || continue
+  echo "Extracting $archive"
+  tar --zstd -xf "$archive"
+done

manifest.json

@@ -0,0 +1,52 @@
+{
+  "description": "Clean public ReGuLaR training mix.",
+  "records": "records.jsonl",
+  "counts_by_source": {
+    "gqa": 245000,
+    "openimage": 50000,
+    "visual_genome": 15000,
+    "psg": 6000,
+    "vsr": 5000,
+    "mgrounding": 5000,
+    "clevr": 25000
+  },
+  "counts_by_dataset": {
+    "gqa": 240000,
+    "openimages_v7": 50000,
+    "blink_proxy_vg_spatial": 10000,
+    "blink_proxy_psg_false": 5000,
+    "blink_proxy_gqa_absence": 5000,
+    "blink_proxy_vsr_true": 5000,
+    "mgrounding_630k": 5000,
+    "blink_proxy_vg_counting": 5000,
+    "blink_proxy_psg_absent_relation_false": 1000,
+    "clevr_v1_referring_relationships": 25000
+  },
+  "skipped_legacy_datasets": {
+    "blink_proxy_sunrgbd_depth": 4000,
+    "blink_proxy_gqa_psg_relation_flip_false": 9109,
+    "blink_proxy_manual_not_facing_false": 401,
+    "blink_proxy_jigsaw": 4000,
+    "blink_proxy_gqa_touching": 336,
+    "blink_proxy_gqa_touching_synth": 25,
+    "blink_proxy_vg_touching": 139
+  },
+  "expected_counts_by_source": {
+    "gqa": 245000,
+    "openimage": 50000,
+    "clevr": 25000,
+    "visual_genome": 15000,
+    "psg": 6000,
+    "mgrounding": 5000,
+    "vsr": 5000
+  },
+  "stats": {
+    "total_records": 351000,
+    "unique_public_images": 140161,
+    "copy_images": true
+  },
+  "parquet": {
+    "total_records": 351000,
+    "total_shards": 15
+  }
+}

records.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:175e2752dd62ecd4b9ced4aeb8731015a9bee69ea0d7d0a5ae12aa2a5c59e819
+size 3232086798

stats.json

@@ -0,0 +1,52 @@
+{
+  "description": "Clean public ReGuLaR training mix.",
+  "records": "records.jsonl",
+  "counts_by_source": {
+    "gqa": 245000,
+    "openimage": 50000,
+    "visual_genome": 15000,
+    "psg": 6000,
+    "vsr": 5000,
+    "mgrounding": 5000,
+    "clevr": 25000
+  },
+  "counts_by_dataset": {
+    "gqa": 240000,
+    "openimages_v7": 50000,
+    "blink_proxy_vg_spatial": 10000,
+    "blink_proxy_psg_false": 5000,
+    "blink_proxy_gqa_absence": 5000,
+    "blink_proxy_vsr_true": 5000,
+    "mgrounding_630k": 5000,
+    "blink_proxy_vg_counting": 5000,
+    "blink_proxy_psg_absent_relation_false": 1000,
+    "clevr_v1_referring_relationships": 25000
+  },
+  "skipped_legacy_datasets": {
+    "blink_proxy_sunrgbd_depth": 4000,
+    "blink_proxy_gqa_psg_relation_flip_false": 9109,
+    "blink_proxy_manual_not_facing_false": 401,
+    "blink_proxy_jigsaw": 4000,
+    "blink_proxy_gqa_touching": 336,
+    "blink_proxy_gqa_touching_synth": 25,
+    "blink_proxy_vg_touching": 139
+  },
+  "expected_counts_by_source": {
+    "gqa": 245000,
+    "openimage": 50000,
+    "clevr": 25000,
+    "visual_genome": 15000,
+    "psg": 6000,
+    "mgrounding": 5000,
+    "vsr": 5000
+  },
+  "stats": {
+    "total_records": 351000,
+    "unique_public_images": 140161,
+    "copy_images": true
+  },
+  "parquet": {
+    "total_records": 351000,
+    "total_shards": 15
+  }
+}