Upload 32 files

README.md

@@ -1,3 +1,190 @@
----
-license: cc-by-sa-4.0
----
+---
+license: cc-by-sa-4.0
+language:
+- en
+pretty_name: laion-coco-hd-molmo-d-7b
+tags:
+- image-text-dataset
+- Molmo
+- captions
+- image captions
+task_categories:
+- text-to-image
+- image-to-text
+- other
+size_categories:
+- 10M<n<100M
+---
+
+# Dataset Card for laion-coco-13m-molmo-d-7b
+
+## Dataset Description
+
+- **Point of Contact:** Caption Emporium
+
+### Dataset Summary
+
+This is 41,409,699 new synthetic captions for the 13,803,233 images found in [laion/laion-coco](https://huggingface.co/datasets/laion/laion-coco). It includes the original captions from that repository as well as new captions. The dataset was filtered to images >= 512px on the short edge.
+
+The long captions were produced using [allenai/Molmo-7B-D-0924](https://huggingface.co/allenai/Molmo-7B-D-0924). Medium and short captions were produced from these captions using [allenai/Llama-3.1-Tulu-3-8B-DPO](https://huggingface.co/allenai/Llama-3.1-Tulu-3-8B-DPO). The dataset was also tagged with [SmilingWolf/wd-eva02-large-tagger-v3](https://huggingface.co/SmilingWolf/wd-eva02-large-tagger-v3) and contains predictions with bounding boxes from [Ultralytics yolov8x-oiv7](https://github.com/ultralytics/ultralytics/blob/main/docs/en/models/yolov8.md). All LLM models were inferenced in 8-bit.
+
+A notable feature of this dataset is double grounding for the captions -- the first set of long captions was grounded with WD tagger tags, and then the second shorter captions were made from the long captions using the alt text for the image.
+
+Mature content was stripped out of the dataset using WD tagger's explicit score.
+
+### Languages
+
+The text is in English, but occasionally text in images in other languages is transcribed.
+
+### Intended Usage
+
+Machine learning research.
+
+### Data Splits
+
+|                                      | train      |
+|--------------------------------------|-----------:|
+| laion-coco-hd-molmo-d-7b             |   13803233 |
+
+## Dataset Creation
+
+### Producing the Dataset
+
+The following prompt was used to produce the captions from the image, VLM, and tags:
+
+```py
+prompt = lambda tags: """
+Please make a detailed caption of this image using as many paragraphs as necessary. If you see text or objects, be sure to describe them in detail along with any other aspects of the foreground and background. As a hint, here is a list of tags for the image that describe visual features.
+
+Hint:
+\`\`\`
+{tag_str}
+\`\`\`
+
+Do not reference the tags in the caption.
+
+Begin caption:
+"""
+```
+
+After producing the caption, they were stripped of prefixes as per [anime-caption-danbooru-2021-sfw-5m-hq](https://huggingface.co/datasets/CaptionEmporium/anime-caption-danbooru-2021-sfw-5m-hq). A pair of medium and short length captions were then produced using [allenai/Llama-3.1-Tulu-3-8B-DPO](https://huggingface.co/allenai/Llama-3.1-Tulu-3-8B-DPO) with the following prompt:
+
+```py
+prompt_shorten = lambda img_prompt, alt_text: f"""
+You have two tasks: prepare a long paragraph length caption and short caption from a long caption and an original caption combined.
+
+## Captions for you to distill
+A. You must determine whether the original caption is relevant, and, if it is, integrate it into the captions you produce.
+
+** Original Caption **
+\`\`\`
+{alt_text}
+\`\`\`
+
+** Detailed Caption **
+\`\`\`
+{img_prompt}
+\`\`\`
+
+## Tasks
+1. Distill it into a long single paragraph that sufficiently describes all details of the image. Remove any redundant lines or descriptions.
+2. Distill it into a single sentence that is a maximum of 25 words for the image. Remove any redundant lines or descriptions.
+
+** Example of Task Output **
+\`\`\`
+**Long Single Paragraph Caption**
+
+An image about removing negative items from a credit report. The image features a structured guide to managing credit reports, divided into three sections: correcting errors (with steps to dispute inaccuracies with credit bureaus and creditors), removing correct entries (explaining "paying for delete" and "goodwill deletion"), and removing negative information from credit reports. A prominent circle highlights the central action of improving credit scores. A document titled "credit report" with a green stripe suggests a form for user engagement, while a person's hand holding a pen implies active participation. The background features a wooden table, adding a professional yet relatable touch. The overall message is about achieving financial freedom by effectively managing one's credit report, with branding elements like "ffg" and "financialfreedom.guru" at the bottom.
+
+**Short Single Sentence Caption**
+
+An image that guides users on how to remove negative items from their credit report, featuring steps to correct errors, remove correct entries, and improve credit scores, with a branded message promoting financial freedom.
+\`\`\`
+
+Please only write the two captions, each on a separate line. Do NOT write either of the original captions verbatim, but instead combine them when creating your new caption.
+"""
+```
+
+This task had a failure rate of 0.0%, so no post-processing cleanup was required.
+
+### Data Instances
+
+The new captions, tags, and bounding boxes are found as the columns `caption_molmo_long`, `caption_molmo_medium`, `caption_molmo_short`, `wd_tagger_tags`, `oi_yolo_pred`.
+
+An example of a row:
+
+```json
+{
+  "caption_molmo_long": "In this serene daytime photograph, we are presented with a beautiful two-story house situated in a lush forested area. The house stands out with its distinctive blue facade and white clear frame windows and doors, creating a charming contrast against the rich greenery of the surrounding trees.\n\nAt the ground level, there's a stone staircase leading up to the house, flanked by a stone wall on one side. The staircase appears to be flanked by several white doors, hinting at the main entrance to the house. The combination of stone and wood elements gives the structure a rustic yet elegant feel.\n\nThe house features a quaint balcony on the second floor, inviting residents to enjoy the surrounding nature. To the right of the house, a white fence can be seen, adding to the property's charm and offering privacy.\n\nThe trees around the house have vibrant green leaves, through which dappled sunlight filters, creating a beautiful interplay of light and shadow. This sunlight casts warm patterns on both the house and the surrounding foliage, enhancing the peaceful atmosphere.\n\nThe blue sky is visible through gaps in the tree canopy, confirming the bright daytime setting. The overall scene depicts a harmonious blend of natural beauty and architectural design, creating a picturesque and tranquil environment for its inhabitant.",
+  "caption_molmo_medium": "A newly constructed home featured in the 2019 showcase of homes. the two-story house stands out with its blue facade and white-framed windows, set against a backdrop of lush greenery. a stone staircase leads up to the main entrance, flanked by a stone wall and white doors. the second floor boasts a charming balcony, while a white fence adds to the property's charm. the vibrant green leaves of the surrounding trees filter dappled sunlight, casting warm patterns on the house and foliage, all under a bright blue sky.",
+  "caption_molmo_short": "Newly constructed home in the 2019 Showcase of Homes, featuring a blue facade, stone staircase, and lush green surroundings.",
+  "wd_tagger_tags": "{\"ratings\":{\"general\":0.99595,\"sensitive\":0.00545,\"questionable\":0.00176,\"explicit\":0.00112},\"character\":{},\"general\":{\"no_humans\":0.98904,\"scenery\":0.98606,\"stairs\":0.9791,\"outdoors\":0.96206,\"tree\":0.95892,\"door\":0.95209,\"house\":0.94795,\"window\":0.94644,\"day\":0.84481,\"building\":0.77799,\"bush\":0.77495,\"grass\":0.76642,\"fence\":0.75981,\"railing\":0.65033,\"sunlight\":0.64838,\"balcony\":0.53143,\"stone_stairs\":0.51395,\"tree_shade\":0.49977,\"plant\":0.48138,\"nature\":0.44787,\"real_world_location\":0.40262,\"stone_wall\":0.36267,\"forest\":0.3612,\"dappled_sunlight\":0.3533}}",
+  "oi_yolo_pred": "[{\"name\":\"House\",\"class\":257,\"confidence\":0.90153,\"box\":{\"x1\":108.40611,\"y1\":0.0,\"x2\":992.15564,\"y2\":671.34094}},{\"name\":\"Window\",\"class\":587,\"confidence\":0.72125,\"box\":{\"x1\":181.93315,\"y1\":335.56635,\"x2\":268.70624,\"y2\":425.39676}},{\"name\":\"Window\",\"class\":587,\"confidence\":0.6585,\"box\":{\"x1\":465.56369,\"y1\":76.65775,\"x2\":594.30737,\"y2\":167.40131}},{\"name\":\"Tree\",\"class\":553,\"confidence\":0.65189,\"box\":{\"x1\":864.8075,\"y1\":0.48331,\"x2\":1080.0,\"y2\":434.53351}},{\"name\":\"Window\",\"class\":587,\"confidence\":0.62657,\"box\":{\"x1\":180.91083,\"y1\":137.73129,\"x2\":268.32281,\"y2\":262.91461}},{\"name\":\"Window\",\"class\":587,\"confidence\":0.62368,\"box\":{\"x1\":769.34808,\"y1\":110.62549,\"x2\":807.96387,\"y2\":234.85358}},{\"name\":\"Window\",\"class\":587,\"confidence\":0.58337,\"box\":{\"x1\":820.40143,\"y1\":140.43805,\"x2\":850.1156,\"y2\":229.35304}},{\"name\":\"Tree\",\"class\":553,\"confidence\":0.42711,\"box\":{\"x1\":0.0,\"y1\":391.80643,\"x2\":192.0901,\"y2\":674.24164}},{\"name\":\"Tree\",\"class\":553,\"confidence\":0.38901,\"box\":{\"x1\":0.01305,\"y1\":0.0,\"x2\":194.14799,\"y2\":673.65466}},{\"name\":\"Porch\",\"class\":401,\"confidence\":0.37016,\"box\":{\"x1\":320.42096,\"y1\":373.21542,\"x2\":858.8009,\"y2\":492.77786}},{\"name\":\"Tree\",\"class\":553,\"confidence\":0.34457,\"box\":{\"x1\":320.70926,\"y1\":0.58194,\"x2\":652.12561,\"y2\":72.78302}},{\"name\":\"Window\",\"class\":587,\"confidence\":0.3105,\"box\":{\"x1\":160.52815,\"y1\":535.77704,\"x2\":260.24438,\"y2\":672.53497}},{\"name\":\"Window\",\"class\":587,\"confidence\":0.28201,\"box\":{\"x1\":483.93518,\"y1\":318.93628,\"x2\":583.94476,\"y2\":441.73453}}]",
+  "TEXT": "2019 Showcase of Homes \u2013 New Construction",
+  "all_captions": [
+    "a blue house with white trim and windows",
+    "the exterior of a house with blue siding",
+    "a blue house with white trim and a porch",
+    "the front view of a house with blue siding",
+    "a blue house with white trim and two story windows",
+    "a blue house with white trim and a black roof",
+    "a blue house with white trim and two story garage",
+    "the front and side view of a house with blue siding"
+  ],
+  "all_similarities": [
+    0.20703125,
+    0.270263671875,
+    0.2232666015625,
+    0.243896484375,
+    0.196044921875,
+    0.222412109375,
+    0.24267578125,
+    0.240478515625
+  ],
+  "similarity": 0.30254724621772766,
+  "pwatermark": 0.10169494152069092,
+  "punsafe": 0.00015485286712646484,
+  "caption": "The exterior of a house with blue siding.",
+  "url": "http://www.kodiakofsaratoga.com/wp-content/uploads/2019/10/IMG_5144-1080x675.png",
+  "key": "131571052",
+  "status": "success",
+  "error_message": null,
+  "width": 1080,
+  "height": 675,
+  "original_width": 1080,
+  "original_height": 675,
+  "exif": "{}",
+  "sha256": "aaaaaa3d4564628e55a994d30b8879cc6fddfff9c5e1204aea8620151b1a547a"
+}
+```
+
+### Discussion of Biases
+
+It is biased to the content of the curated dataset along with the training data of Molmo and the taggers.
+
+### Known Limitations
+
+- Captions are not manually verified for correctness, so there may be errors.
+
+## Additional Information
+
+### Dataset Curators
+
+Caption Emporium, [LAION](https://huggingface.co/laion).
+
+### Licensing Information
+
+The dataset is available under the [Creative Commons ShareAlike (CC BY-SA 4.0)](https://creativecommons.org/licenses/by-sa/4.0/legalcode.en).
+
+### Citation Information
+
+```
+@misc{laion-coco-hd-molmo-d-7b,
+  author = { Caption Emporium },
+  title = {laion-coco-hd-molmo-d-7b},
+  year = {2025},
+  publisher = {Huggingface},
+  journal = {Huggingface repository},
+  howpublished = {\url{https://huggingface.co/datasets/CaptionEmporium/laion-coco-hd-molmo-d-7b}},
+}
+```

train/part_000.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5eec3b55fd2e03d63aa4e99e0607b8a8546456cc44015a3f6930c7b420a25486
+size 1073540017

train/part_001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6737dfb8e94a98569a45fd03c73de92a0740a5816ce3008a0eece0b685f28b43
+size 1073267858

train/part_002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22c7f044db84f6503a63eb7b3db163e135660bde808fc28405b9d55ba1df0af6
+size 1073696007

train/part_003.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70e02c587beeb178645828ffa431130781daf34080ee9b902d43eafc0362a87b
+size 1073262708

train/part_004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb5ed947f47cc093844cde2a8f7bbdb61a86f5268c2876ae2d6cbffb357d1526
+size 1073803508

train/part_005.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d310031ebbacb93b2a181bfa592f4c81bd14e3c091f330ca7f35b3096b4f5f8
+size 1073300815

train/part_006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd5aaadfb6c700586ee98b500062042ff96bcd9e7e252c9d73f454a18181bf2d
+size 1073309362

train/part_007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:283cc2e9a8676ef1a9c543807dbecaf193cc3690b7c753f15a445a91501991d3
+size 1073829866

train/part_008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:855fa3b57fb82bdce4bf7ab9763cb3421c9e99bf42d98911cc8588a94918367d
+size 1073725803

train/part_009.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0eb37939b8e48ef047a5e0f4ab280b8eb00f4a459020d6af461aa9aca2da6ab
+size 1073411602

train/part_010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:920ee0b71cce639ea1861f91883827a9a89b5986e1e90bd1287c287985890994
+size 1073274831

train/part_011.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa7f8af89e82145839cc7ed1ac4f83bdabe30bd3ce169c641ef2ea7e908a4b60
+size 1073664289

train/part_012.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70ad747a8f95cb50c24b55d244ae2e74af6d38adb71798d6faa1f9cb39a22a02
+size 1073009219

train/part_013.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3daf8bd63d0ec0440f5839d39f63ffc351d9ed4179946246ae3040ae464b8acf
+size 1073041191

train/part_014.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c14a3b888055af1646a5abc6a135683d1aa55b5309b5b877f9b80f21bd903016
+size 1073378738

train/part_015.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f79defb514746813ba61a96e1a311a374095183afd2df7b997658030d225357
+size 1073619209

train/part_016.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a87f7fefd35f1d88b8132c72887d6eb596f35c93dd8ccb4b7dec5735387cdba5
+size 1073724238

train/part_017.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e44053909c78cbfae215e196839c545d59b6c9e2cdd04b45cf6da2f0eace0ba6
+size 1072874620

train/part_018.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b17fd88b8308e39fa4eaef782f0232634507adbb044b4b9b423eceef2d081fc
+size 1073741573

train/part_019.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d8d3731e67b8c282fda7c0dcb26813b2890e4c2f883a390eba9442ba04b8629
+size 1073711348

train/part_020.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06d83ebebe6a70b7ad24094303593abf0ba4164e632f079efadb498244a11379
+size 1073883984

train/part_021.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ac087884724760e2753a6b362e9e91ce3023887d5f2ded17fe60ea53008cf1f
+size 1072686079

train/part_022.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f197a585100fde6f959174ddc0572493e2f4638bccd48ace02317b160f76ae1
+size 1073832601

train/part_023.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:daba37c5e0ca29db9278d4341c86cd2c8cdc50cc2e9818cd8e61f83bf4b8ce73
+size 1072947714

train/part_024.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eca1c2f4c12abc900c24487777086f5f804edbd25a51106aa5618e464b1eac00
+size 1073505304

train/part_025.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac9221a3a0d6bd8219faf8bddcf1636ffe15a6c1e0b22be09714a15eff2b2534
+size 1073490755

train/part_026.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:110b52cd594e310d0fadcbb237353a3203c0800bb7acca29a9986207e6b8efe9
+size 1073329952

train/part_027.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a0a135b76a5f35111408195cf95de636f65d6798406343ec2591aba71086f18
+size 1073415789

train/part_028.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bedae8128e4805bdb208f0f3bc344a80dbc8f342b3d6de6d02d9d43a44d9719
+size 1073011935

train/part_029.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7307fb1fa8ef35bb8cd22f40b91dec921d836fa881a6099b5fdaca26baaf3e85
+size 1073963838

train/part_030.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01db41527550e2a16265b22ebf61263b7685103686bff67bbbb2bc9ab25bb0dd
+size 326596721