data-archetype/bg_photo_concepts_bucketed_1024

bghira_photo_concepts_bucketed_1024

Edit this section before uploading:

• Title: bghira_photo_concepts_bucketed_1024
• Description: (add a short description / intended use)

---

Technical details

This repository contains a bucketed-shards export (uncompressed TAR shards).

Format

• Format: bucketed_shards_v2
• Created: 2026-04-30T11:47:20.761643+00:00
• Export ID: export-2026-04-30T11:47:20.761643+00:00
• Manifest: manifest.json
• Image mode: reencode_jpeg

Directory layout:

• manifest.json (global metadata + per-bucket shard listing)
• buckets/<bucket_id>/shard-*.tar

Each TAR shard contains 3 files per sample:

• <key>.jpg (JPEG bytes; either re-encoded RGB JPEG or source JPEG passthrough depending on image_mode)
• <key>.txt (caption text, UTF-8, newline-terminated)
• <key>.json (per-sample metadata: w, h, jpeg, image_mode, caption_variant, caption_selector_index, caption_source_id)

Image preprocessing

If image_mode=reencode_jpeg, images are processed deterministically per-sample:

• EXIF transpose, convert to RGB
• Cover-resize using torch CPU bicubic interpolation with antialiasing (mode=bicubic, antialias=True)
• Never upsample: samples that would require upscaling are skipped (too_small_policy=drop)
• Crop to the bucket target size (crop_strategy=corner, allowed corners [2, 3])

Resize/crop details:

• Cover scale is scale = max(target_w / src_w, target_h / src_h); if scale > 1, the sample is skipped.
• After resize, a crop box is chosen deterministically from the sample key (sha256 of image_id).
  • Corner strategy chooses a corner from allowed_corners where 0=TL, 1=TR, 2=BL, 3=BR (optional small jitter for corner_jitter).

JPEG encoding:

• quality 95
• subsampling policy adaptive_scale (adaptive threshold 0.85)

If image_mode=passthrough_jpeg, the exporter stores the source file bytes as-is (no EXIF transpose / resize / crop / re-encode). Buckets and w/h metadata still refer to the planned target size for that bucket (not necessarily the encoded JPEG dimensions). Loaders should decode the JPEG bytes, apply EXIF orientation if desired, then do resize/crop at load time.

Buckets / resolutions

• Buckets follow SDXL-style proto buckets defined at a 1024×1024 base.
• Base resolution(s): [1024, 768, 512]
• In single-res exports, bucket_id is the proto (1024-base) bucket, e.g. p1024x1024.
• In multi-res exports, buckets are namespaced by base resolution: r<base>_<proto>, e.g. r512_p1024x1024.
• The actual target resolution for each bucket (scaled by the per-bucket base resolution and divisible=32) is stored in:
  • manifest.json → buckets[<bucket_id>].scaled.w/h (and base_resolution)
  • each sample’s <key>.json → w/h

Bucket IDs (preview): r1024_p1024x1024, r1024_p1024x960, r1024_p1088x896, r1024_p1088x960, r1024_p1152x832, r1024_p1152x896, r1024_p1216x832, r1024_p1280x768, r1024_p1344x704, r1024_p1344x768, r1024_p1408x704, r1024_p1472x704, r1024_p1536x640, r1024_p1600x640, r1024_p1664x576, r1024_p1728x576, r1024_p1792x576, r1024_p1856x512, r1024_p1984x512, r1024_p2048x512, … (+18 more)

Bucket distribution:

bucket_id	target_w×h	aspect	count
r1024_p832x1216	832×1216	0.684	244,885
r1024_p1216x832	1216×832	1.462	187,336
r1024_p832x1152	832×1152	0.722	41,645
r1024_p896x1088	896×1088	0.824	19,013
r1024_p1152x832	1152×832	1.385	16,881
r1024_p768x1344	768×1344	0.571	14,895
r1024_p1344x768	1344×768	1.750	12,149
r1024_p896x1152	896×1152	0.778	6,931
r1024_p1024x1024	1024×1024	1.000	4,742
r1024_p1280x768	1280×768	1.667	4,648
r1024_p768x1280	768×1280	0.600	3,732
r1024_p1088x896	1088×896	1.214	2,953
r1024_p1152x896	1152×896	1.286	2,809
r1024_p960x1088	960×1088	0.882	1,074
r1024_p960x1024	960×1024	0.938	756
r1024_p1088x960	1088×960	1.133	737
r1024_p1024x960	1024×960	1.067	668
r1024_p1344x704	1344×704	1.909	627
r1024_p1408x704	1408×704	2.000	272
r1024_p704x1408	704×1408	0.500	242
r1024_p1472x704	1472×704	2.091	210
r1024_p1536x640	1536×640	2.400	117
r1024_p704x1472	704×1472	0.478	102
r1024_p1600x640	1600×640	2.500	28
r1024_p1664x576	1664×576	2.889	23
r1024_p1728x576	1728×576	3.000	13
r1024_p640x1536	640×1536	0.417	10
r1024_p1792x576	1792×576	3.111	4
r1024_p1856x512	1856×512	3.625	4
r768_p1216x832	896×608	1.474	3
r1024_p2048x512	2048×512	4.000	2
r1024_p576x1664	576×1664	0.346	2
r1024_p640x1600	640×1600	0.400	2
r1024_p1984x512	1984×512	3.875	1
r1024_p512x2048	512×2048	0.250	1
r512_p1216x832	608×416	1.462	1
r512_p2048x512	1024×256	4.000	1
r768_p832x1216	608×896	0.679	1

Caption selection (waterfall)

Captions are selected from dataset.sqlite using the first matching selector (highest priority wins). Within the same selector, the newest caption source is preferred.

Caption provenance:

• Per-sample <key>.json includes caption_source_id (int, from dataset.sqlite).
• manifest.json includes a caption_sources table mapping caption_source_id → backend/model/created_at plus prompt hashes (not prompt text).

Caption sources used:

caption_source_id	backend	model	created_at	system_prompt_sha256	user_prompt_sha256
1	legacy	bghira_photo_concepts_core	1767891404	e3b0c44298fc…	e3b0c44298fc…
6	openrouter	google/gemini-3-flash-preview	1777545618	256027f6c923…	6ed671b1ecec…
7	openrouter	mistralai/mistral-medium-3.1	1777548969	256027f6c923…	6ed671b1ecec…

Caption priority (waterfall) + planned usage:

selector_index	variant	backend	model	planned_images
0	caption_gemini_3_flash_preview			2,836
1	caption_gemini			564,487
2	caption_mistral_medium_3_1			197

Available caption variants (top 30):

selected	variant	images_with_ok_caption
✓	caption_gemini	565,600
✓	caption_gemini_3_flash_preview	2,836
✓	caption_mistral_medium_3_1	197

Missing caption policy: drop

Export summary

• images_seen: 567,520
• images_exported: 567,520
• skipped_no_caption: 0
• skipped_too_small: 0
• decode_errors: 0
• encode_errors: 0

Efficient loading

Recommended

Treat this as a webdataset-style collection of tar shards:

• Prefer sequential reads of tar files for throughput.
• Shuffle at the shard level (and optionally within-shard) for good randomness without expensive random I/O.
• Use manifest.json to list buckets and shards.

Python (webdataset)

import webdataset as wds

urls = "buckets/*/shard-*.tar"  # glob; adjust if you want a single bucket only
ds = (
    wds.WebDataset(urls)
    .decode("pil")            # decodes .jpg to PIL.Image
    .to_tuple("jpg", "txt", "json")
)
for jpg, caption, meta in ds:
    ...

Python (tarfile, no extra deps)

import io, json, tarfile
from pathlib import Path

tar_path = next(Path("buckets").rglob("shard-*.tar"))
with tarfile.open(tar_path, "r") as tf:
    members = tf.getmembers()
    for m in members:
        if not m.name.endswith(".txt"):
            continue
        key = m.name[:-4]
        caption = tf.extractfile(m).read().decode("utf-8").strip()
        meta = json.loads(tf.extractfile(tf.getmember(key + ".json")).read().decode("utf-8"))
        jpg_bytes = tf.extractfile(tf.getmember(key + ".jpg")).read()
        ...