wavetao2010 commited on
Commit
c19e39d
·
verified ·
1 Parent(s): afe4443

Upload 2 files

Browse files
Files changed (2) hide show
  1. README.md +80 -41
  2. croissant.json +2286 -0
README.md CHANGED
@@ -28,26 +28,52 @@ tags:
28
  ---
29
 
30
  # OpenVTON
31
- A large-scale virtual try-on dataset containing ~100K clothing image pairs with garment masks.
32
 
33
- ## Dataset Structure
34
 
35
- Each sample contains:
36
- - **source**: Garment image (clothing item)
37
- - **mask**: Garment segmentation mask
38
- - **target**: Person wearing the garment (ground truth)
39
- - **caption**: Text description of the clothing
40
- - **category**: Clothing category (e.g., pants, jeans, shirt)
 
 
 
 
 
 
 
 
 
 
 
41
 
42
  ## Splits
43
 
44
- | Split | Samples |
45
- |-------|---------|
46
  | Train | 89,927 |
47
  | Validation | 4,989 |
48
  | Test | 5,009 |
49
  | **Total** | **99,925** |
50
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
51
  ## Usage
52
 
53
  ```python
@@ -55,55 +81,68 @@ from datasets import load_dataset
55
 
56
  dataset = load_dataset("RenxingIntelligence/OpenVTON")
57
  sample = dataset["train"][0]
58
- sample["source"].show() # garment image
59
- sample["mask"].show() # segmentation mask
60
- sample["target"].show() # person wearing garment
61
  print(sample["caption"])
62
  print(sample["category"])
63
  ```
64
- ## Benchmark and Paper
65
 
66
- This dataset is part of **OpenVTON-Bench**, a large-scale benchmark designed for the systematic evaluation of controllable virtual try-on (VTON) models.
 
 
 
 
 
 
 
 
 
67
 
68
- **OpenVTON-Bench** is introduced in our paper:
69
 
70
- > **OpenVTON-Bench: A Large-Scale Benchmark for Controllable Virtual Try-On**
71
- > 📄 Paper: [https://arxiv.org/abs/2601.22725](https://arxiv.org/abs/2601.22725)
72
- > 💻 Code: [https://github.com/RenxingIntelligence/OpenVTON-Bench](https://github.com/RenxingIntelligence/OpenVTON-Bench)
73
 
74
- OpenVTON-Bench provides a standardized evaluation protocol for modern diffusion-based and transformer-based virtual try-on systems, enabling fair and reproducible comparison across different architectures.
75
 
76
- ---
77
 
78
- ## About OpenVTON-Bench
79
 
80
- **OpenVTON-Bench** is a **large-scale, high-resolution benchmark** designed for the **systematic evaluation of controllable virtual try-on models**.
81
 
82
- Unlike existing datasets and evaluation protocols that struggle with texture details and semantic consistency, OpenVTON-Bench provides:
83
 
84
- * 🖼️ **~100K Image Pairs** with resolutions up to **1536×1536**, enabling evaluation of fine-grained texture generation.
85
- * 🏷️ **Fine-Grained Taxonomy** covering **20 garment categories** for balanced semantic evaluation.
86
- * 📐 **Multi-Level Automated Evaluation**, including:
87
 
88
- * Pixel fidelity
89
- * Garment consistency
90
- * Semantic realism
91
 
92
- This benchmark enables **fair, reproducible, and scalable comparison** across modern virtual try-on systems.
93
 
94
- ---
 
 
 
 
 
 
 
 
 
 
95
 
96
  ## Citation
97
- If you use this dataset or the benchmark in your research, please cite:
 
 
98
  ```bibtex
99
  @misc{li2026openvtonbenchlargescalehighresolutionbenchmark,
100
- title={OpenVTON-Bench: A Large-Scale High-Resolution Benchmark for Controllable Virtual Try-On Evaluation},
101
- author={Jin Li and Tao Chen and Shuai Jiang and Weijie Wang and Jingwen Luo and Chenhui Wu},
102
- year={2026},
103
- eprint={2601.22725},
104
- archivePrefix={arXiv},
105
- primaryClass={cs.CV},
106
- url={https://arxiv.org/abs/2601.22725},
107
  }
108
  ```
109
-
 
28
  ---
29
 
30
  # OpenVTON
 
31
 
32
+ OpenVTON is the dataset released with **OpenVTON-Bench: A Large-Scale High-Resolution Benchmark for Controllable Virtual Try-On Evaluation**. It contains **99,925 garment-person pairs**, with masking images and garment captions, for controllable virtual try-on research and evaluation.
33
 
34
+ - [Paper](https://arxiv.org/abs/2601.22725)
35
+ - [Evaluation code and benchmark documentation](https://github.com/RenxingIntelligence/OpenVTON-Bench)
36
+ - [Croissant metadata with Responsible AI information](croissant.json)
37
+
38
+ ## Dataset structure
39
+
40
+ Each row contains the following five fields:
41
+
42
+ | Field | Type | Description |
43
+ |:---|:---|:---|
44
+ | `source` | Image | Reference garment image |
45
+ | `mask` | Image | Person image with the target garment region blacked out; masked-person conditioning input |
46
+ | `target` | Image | Person wearing the garment; target image for the paired evaluation |
47
+ | `caption` | String | Garment description generated with Gemini 2.0 Flash |
48
+ | `category` | String | One of 14 coarse labels extracted by keyword matching over captions |
49
+
50
+ The `category` column contains **caption-derived metadata labels**. It does not store the DINOv3 visual cluster assignment.
51
 
52
  ## Splits
53
 
54
+ | Split | Pairs |
55
+ |:---|---:|
56
  | Train | 89,927 |
57
  | Validation | 4,989 |
58
  | Test | 5,009 |
59
  | **Total** | **99,925** |
60
 
61
+ The release consists of 18 training, one validation, and two test Parquet shards. Split sizes and statistics here describe the complete release, rather than a dataset-viewer preview.
62
+
63
+ ## Visual sampling clusters and caption labels
64
+
65
+ Final sampling uses **20 visual clusters (C0-C19)** obtained by applying **K-Means with K=20 to DINOv3 garment embeddings**. These are visual sampling groups, not 20 manually defined garment categories.
66
+
67
+ The separate `category` field contains **14 coarse labels** obtained by deterministic keyword matching over Gemini captions, with first-match precedence and the merges `tshirt -> top`, `knitwear -> sweater`, and `suit -> jacket`:
68
+
69
+ `blazer`, `cardigan`, `coat`, `dress`, `hoodie`, `jacket`, `jeans`, `pants`, `shirt`, `shorts`, `skirt`, `sweater`, `top`, and `vest`.
70
+
71
+ One caption label can span several visual clusters. Approximately balanced sampling across clusters therefore does not imply uniform caption-label frequencies or demographic balance. The labels are not a manually verified garment taxonomy.
72
+
73
+ Nineteen clusters contain 5,012-5,088 pairs each. C13 contains all 3,340 available pairs from its source cluster. Sampling is approximately balanced across clusters, subject to source availability, and the splits preserve these cluster proportions.
74
+
75
+ The [project README](https://github.com/RenxingIntelligence/OpenVTON-Bench#data-splits-and-sampling-clusters) provides the complete **14 x 20 label-by-cluster cross-tabulation**, **C0-C19 counts by split**, and **dataset taxonomy comparison**. For example, the full-release frequencies of `pants` and `shorts` are 12.01% and 8.97%, respectively.
76
+
77
  ## Usage
78
 
79
  ```python
 
81
 
82
  dataset = load_dataset("RenxingIntelligence/OpenVTON")
83
  sample = dataset["train"][0]
84
+ sample["source"].show() # reference garment
85
+ sample["mask"].show() # person image with the garment region blacked out
86
+ sample["target"].show() # target person image
87
  print(sample["caption"])
88
  print(sample["category"])
89
  ```
 
90
 
91
+ To use the data revision described in the accompanying Croissant metadata:
92
+
93
+ ```python
94
+ dataset = load_dataset(
95
+ "RenxingIntelligence/OpenVTON",
96
+ revision="afe4443ab7961cabe00ec3122809c89111039cde",
97
+ )
98
+ ```
99
+
100
+ ## Construction and provenance
101
 
102
+ The dataset combines curated garment-person pairs from the [IG Pair dataset associated with IMAGDressing-v1](https://github.com/muzishen/IMAGDressing) and publicly available e-commerce imagery. Quality and resolution filtering, including aesthetic scoring and face detection, were followed by human checks of garment-person correspondence and image completeness.
103
 
104
+ DINOv3 garment embeddings and K-Means define the visual sampling clusters. Gemini 2.0 Flash generates dense garment descriptions using upper-body/lower-body routing and structured prompts. GroundingDINO and SAM3 are used for garment localization and masking; Qwen-VL-Plus supplies post-hoc descriptions for mask prompting. These descriptions do not define cluster membership. The [published prompts](https://github.com/RenxingIntelligence/OpenVTON-Bench/blob/main/prompts.md) and paper Section 3 / Appendix A document the construction pipeline.
 
 
105
 
106
+ **Synthetic and transformed components:** the release combines photographs with generated captions and algorithmically transformed images, including masking and generative facial anonymization. It is not a wholly synthetic-photo dataset. Accordingly, the Croissant metadata declares `rai:hasSyntheticData: true` and describes the processing activities.
107
 
108
+ ## Intended use, limitations, and responsible use
109
 
110
+ OpenVTON supports non-commercial research on controllable virtual try-on, including model evaluation and adaptation. The accompanying benchmark combines semantic, representation, and pixel-level metrics. Its human-study results concern **model-ranking agreement under a shared visual-quality rubric**. The benchmark does not establish physical fit, sizing, garment grading, or material physics.
111
 
112
+ E-commerce and social-media source selection, filtering, and automated annotation may introduce selection or label biases. Extreme poses and complex multi-layer occlusions may be underrepresented. Captions and masks can contain errors, and keyword-matched labels may reflect incidental garment mentions. Demographic representativeness and parity have not been established; the released schema has no demographic annotation fields.
113
 
114
+ Images depict people and may convey apparent personal attributes. Facial regions underwent generative anonymization before release, as described in the paper. Users should not use the dataset for biometric identification, face recognition, surveillance, or attempts to recover personal identity. Potential misuse also includes deceptive imagery, unauthorized person-image editing, and unrealistic body portrayals. The paper describes a takedown mechanism for individuals and rights holders.
115
 
116
+ ## Benchmark configurations
 
 
117
 
118
+ The [baseline configurations](https://github.com/RenxingIntelligence/OpenVTON-Bench#baseline-configurations) document the reported model identifiers, checkpoints, inference settings, sampling steps, guidance scales, and seeds. The benchmark applies common aspect-ratio-preserving resizing and centered white padding for evaluation, independently of native inference dimensions. API-based results refer to fixed generated outputs.
 
 
119
 
120
+ ## Croissant metadata
121
 
122
+ Use the checked-in **[croissant.json](croissant.json)** for the camera-ready metadata submission. It includes both core Croissant fields and the Responsible AI fields requested by NeurIPS 2026, including data provenance, intended uses, limitations, biases, personal-information considerations, social impact, and synthetic components.
123
+
124
+ The metadata describes all **21 original Parquet shards** through direct, revision-pinned download URLs and the SHA-256 checksums published by Hugging Face. Each record set represents one physical shard; concatenate record sets sharing a split prefix to obtain the corresponding train, validation, or test split. The metadata's version `1.0.0+hf.afe4443ab7961cabe00ec3122809c89111039cde` identifies this documented data snapshot; it is not a claim that a tag with that name exists on Hugging Face.
125
+
126
+ Hugging Face also offers automatically generated Croissant metadata. The checked-in file adds dataset-specific RAI information and should be updated whenever data shards change.
127
+
128
+ ## License
129
+
130
+ This dataset is released under [Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)](https://creativecommons.org/licenses/by-nc/4.0/). Sharing and adaptation for non-commercial purposes are subject to attribution and the other license conditions. Original asset terms and third-party rights continue to apply. Recommendations for responsible use are separate from the license terms.
131
+
132
+ The benchmark code has a separate license documented in its [repository](https://github.com/RenxingIntelligence/OpenVTON-Bench).
133
 
134
  ## Citation
135
+
136
+ If you use this dataset or benchmark, please cite the paper:
137
+
138
  ```bibtex
139
  @misc{li2026openvtonbenchlargescalehighresolutionbenchmark,
140
+ title={OpenVTON-Bench: A Large-Scale High-Resolution Benchmark for Controllable Virtual Try-On Evaluation},
141
+ author={Jin Li and Tao Chen and Shuai Jiang and Weijie Wang and Jingwen Luo and Chenhui Wu},
142
+ year={2026},
143
+ eprint={2601.22725},
144
+ archivePrefix={arXiv},
145
+ primaryClass={cs.CV},
146
+ url={https://arxiv.org/abs/2601.22725}
147
  }
148
  ```
 
croissant.json ADDED
@@ -0,0 +1,2286 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "@context": {
3
+ "@language": "en",
4
+ "@vocab": "https://schema.org/",
5
+ "citeAs": "cr:citeAs",
6
+ "column": "cr:column",
7
+ "conformsTo": "dct:conformsTo",
8
+ "cr": "http://mlcommons.org/croissant/",
9
+ "rai": "http://mlcommons.org/croissant/RAI/",
10
+ "data": {
11
+ "@id": "cr:data",
12
+ "@type": "@json"
13
+ },
14
+ "dataType": {
15
+ "@id": "cr:dataType",
16
+ "@type": "@vocab"
17
+ },
18
+ "dct": "http://purl.org/dc/terms/",
19
+ "equivalentProperty": "cr:equivalentProperty",
20
+ "examples": {
21
+ "@id": "cr:examples",
22
+ "@type": "@json"
23
+ },
24
+ "extract": "cr:extract",
25
+ "field": "cr:field",
26
+ "fileProperty": "cr:fileProperty",
27
+ "fileObject": "cr:fileObject",
28
+ "fileSet": "cr:fileSet",
29
+ "format": "cr:format",
30
+ "includes": "cr:includes",
31
+ "isLiveDataset": "cr:isLiveDataset",
32
+ "jsonPath": "cr:jsonPath",
33
+ "key": "cr:key",
34
+ "md5": "cr:md5",
35
+ "parentField": "cr:parentField",
36
+ "path": "cr:path",
37
+ "recordSet": "cr:recordSet",
38
+ "references": "cr:references",
39
+ "regex": "cr:regex",
40
+ "repeated": "cr:repeated",
41
+ "replace": "cr:replace",
42
+ "samplingRate": "cr:samplingRate",
43
+ "sc": "https://schema.org/",
44
+ "separator": "cr:separator",
45
+ "source": "cr:source",
46
+ "subField": "cr:subField",
47
+ "transform": "cr:transform",
48
+ "prov": "http://www.w3.org/ns/prov#"
49
+ },
50
+ "@type": "sc:Dataset",
51
+ "name": "OpenVTON",
52
+ "alternateName": [
53
+ "OpenVTON-Bench",
54
+ "RenxingIntelligence/OpenVTON"
55
+ ],
56
+ "description": "OpenVTON-Bench contains 99,925 garment-person image pairs with associated masking images and captions: 89,927 training, 4,989 validation, and 5,009 test pairs. Sampling uses 20 visual clusters (C0-C19) from K-Means over DINOv3 garment embeddings. The category column separately contains 14 coarse labels extracted from Gemini captions. This Croissant file describes every published Parquet shard at a fixed data revision; record sets with the same split prefix are concatenated to obtain that split.",
57
+ "url": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON",
58
+ "license": "https://creativecommons.org/licenses/by-nc/4.0/",
59
+ "conformsTo": "http://mlcommons.org/croissant/1.1",
60
+ "version": "1.0.0+hf.afe4443ab7961cabe00ec3122809c89111039cde",
61
+ "cr:sdVersion": "1.0.0",
62
+ "dateCreated": "2026-03-16T06:18:57Z",
63
+ "dateModified": "2026-10-09",
64
+ "creator": {
65
+ "@type": "sc:Organization",
66
+ "name": "RenxingIntelligence",
67
+ "url": "https://huggingface.co/RenxingIntelligence"
68
+ },
69
+ "citeAs": "Jin Li et al. OpenVTON-Bench: A Large-Scale High-Resolution Benchmark for Controllable Virtual Try-On Evaluation. arXiv:2601.22725, 2026. https://arxiv.org/abs/2601.22725",
70
+ "keywords": [
71
+ "virtual try-on",
72
+ "benchmark",
73
+ "image-to-image",
74
+ "DINOv3",
75
+ "visual sampling clusters"
76
+ ],
77
+ "rai:dataLimitations": [
78
+ "The benchmark evaluates 2D visual try-on quality. It does not establish physical fit, sizing, garment grading, or material physics.",
79
+ "Extreme poses, complex multi-layer occlusions, and imaging conditions outside the source distribution may be underrepresented.",
80
+ "Caption-derived labels and automated masks may contain errors. Approximate balancing of visual clusters does not imply balanced caption labels or demographic groups."
81
+ ],
82
+ "rai:dataBiases": [
83
+ "Source selection from e-commerce and social-media imagery, aesthetic and face-based filtering, and human screening may favor particular clothing styles, presentation conditions, and visible-body configurations.",
84
+ "Gemini captions and DINOv3 representations may introduce model-dependent biases. Keyword matching may label an incidental garment mentioned in a caption.",
85
+ "Population representativeness and demographic parity have not been established; the release does not contain demographic annotations."
86
+ ],
87
+ "rai:personalSensitiveInformation": [
88
+ "The images depict people and may convey apparent personal attributes. Facial regions underwent generative anonymization before release, as described in Section 3.1 of the paper.",
89
+ "The released columns are source, mask, target, caption, and category; no participant identity or human-rating records are included in these Parquet files.",
90
+ "Users should not use this dataset for identity recovery, biometric identification, face recognition, or surveillance."
91
+ ],
92
+ "rai:dataUseCases": [
93
+ "Non-commercial research on controllable virtual try-on, including model evaluation and adaptation using paired garment/person images and masked conditioning inputs.",
94
+ "The paper supports model-ranking agreement with human evaluation under a shared five-dimensional visual-quality rubric. This is not a validation of physical fit or unrestricted human-preference prediction.",
95
+ "Commercial use is outside CC BY-NC 4.0 authorization. Original asset terms and third-party rights continue to apply."
96
+ ],
97
+ "rai:dataSocialImpact": "The benchmark supports transparent comparisons and analysis of garment fidelity, identity/body consistency, and background preservation. Potential misuse includes deceptive synthetic imagery, unauthorized editing of person images, identity inference, and unrealistic body portrayals. Mitigations described in the paper include facial anonymization, non-commercial licensing, and a takedown mechanism for individuals or rights holders.",
98
+ "rai:hasSyntheticData": true,
99
+ "rai:dataCollection": "Curated garment-person pairs from the IG Pair dataset associated with IMAGDressing-v1 were combined with publicly available e-commerce imagery. Human screening checked garment-person correspondence and image completeness; the paper describes the source collection and filtering in Section 3 and Appendix A.",
100
+ "rai:dataPreprocessingProtocol": [
101
+ "Resolution and quality filtering, including aesthetic scoring and face detection, preceded final sampling.",
102
+ "DINOv3 garment embeddings were clustered using K-Means with K=20. Final sampling approximately balances these clusters subject to source availability; the released splits contain 89,927/4,989/5,009 pairs.",
103
+ "Facial regions were generatively anonymized. GroundingDINO and SAM3 were used for garment localization and masking. The benchmark includes original photographs together with transformed images and generated annotations; it is not a wholly synthetic-photo dataset."
104
+ ],
105
+ "rai:dataAnnotationProtocol": [
106
+ "Human verification checked garment-person matches and visible body completeness.",
107
+ "Gemini 2.0 Flash generated garment captions using upper-body/lower-body routing and structured prompts. The released category column was derived by deterministic first-match keyword matching, with tshirt mapped to top, knitwear to sweater, and suit to jacket.",
108
+ "Qwen-VL-Plus suggested post-hoc garment descriptions for mask prompting; these descriptions did not determine visual-cluster membership. Existing prompts are documented at https://github.com/RenxingIntelligence/OpenVTON-Bench/blob/main/prompts.md."
109
+ ],
110
+ "rai:machineAnnotationTools": [
111
+ "Gemini 2.0 Flash",
112
+ "DINOv3",
113
+ "Qwen-VL-Plus",
114
+ "GroundingDINO",
115
+ "SAM3"
116
+ ],
117
+ "rai:dataReleaseMaintenancePlan": "The dataset is publicly hosted on Hugging Face; benchmark documentation and implementation are maintained at https://github.com/RenxingIntelligence/OpenVTON-Bench. The paper describes a takedown mechanism for individuals and copyright holders. This metadata is pinned to the data revision named in version and should be regenerated if data shards change.",
118
+ "prov:wasDerivedFrom": [
119
+ {
120
+ "@id": "https://github.com/muzishen/IMAGDressing"
121
+ }
122
+ ],
123
+ "prov:wasGeneratedBy": [
124
+ {
125
+ "@id": "#collection-and-screening",
126
+ "@type": "prov:Activity",
127
+ "sc:description": "Integration of curated IG Pair garment-person pairs and publicly available e-commerce imagery; quality filtering and human pair verification. See paper Section 3.1-3.2 and Appendix A.2."
128
+ },
129
+ {
130
+ "@id": "#visual-cluster-sampling",
131
+ "@type": "prov:Activity",
132
+ "sc:description": "DINOv3 garment embedding extraction, K-Means K=20, and cluster-stratified sampling/splitting. Complete counts: https://github.com/RenxingIntelligence/OpenVTON-Bench#data-splits-and-sampling-clusters."
133
+ },
134
+ {
135
+ "@id": "#annotation-and-transformation",
136
+ "@type": "prov:Activity",
137
+ "sc:description": "Generative face anonymization, Gemini caption generation, deterministic caption-label extraction, and GroundingDINO/SAM3 garment masking. See paper Section 3 and Appendix A; published prompts: https://github.com/RenxingIntelligence/OpenVTON-Bench/blob/main/prompts.md."
138
+ }
139
+ ],
140
+ "distribution": [
141
+ {
142
+ "@type": "cr:FileObject",
143
+ "@id": "file-test-00000-of-00002.parquet",
144
+ "name": "test-00000-of-00002.parquet",
145
+ "description": "Published test split shard.",
146
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/test/test-00000-of-00002.parquet",
147
+ "encodingFormat": "application/x-parquet",
148
+ "contentSize": "2553937933",
149
+ "sha256": "bdd1e5855b6b96afd948a4575397926fdc6efbb410c2757a7d4145375498a557"
150
+ },
151
+ {
152
+ "@type": "cr:FileObject",
153
+ "@id": "file-test-00001-of-00002.parquet",
154
+ "name": "test-00001-of-00002.parquet",
155
+ "description": "Published test split shard.",
156
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/test/test-00001-of-00002.parquet",
157
+ "encodingFormat": "application/x-parquet",
158
+ "contentSize": "4191182",
159
+ "sha256": "b6b979b5104d6542772240e55cc586403f18b9d847d928249f6c7aa03d7b2911"
160
+ },
161
+ {
162
+ "@type": "cr:FileObject",
163
+ "@id": "file-train-00000-of-00018.parquet",
164
+ "name": "train-00000-of-00018.parquet",
165
+ "description": "Published train split shard.",
166
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00000-of-00018.parquet",
167
+ "encodingFormat": "application/x-parquet",
168
+ "contentSize": "2223983159",
169
+ "sha256": "557b02f5121d20e48bf48c28c6face30a2c52bce3fb0726a01533ce581e92db2"
170
+ },
171
+ {
172
+ "@type": "cr:FileObject",
173
+ "@id": "file-train-00001-of-00018.parquet",
174
+ "name": "train-00001-of-00018.parquet",
175
+ "description": "Published train split shard.",
176
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00001-of-00018.parquet",
177
+ "encodingFormat": "application/x-parquet",
178
+ "contentSize": "2775679372",
179
+ "sha256": "eccdf6cf569dabb6536d452376469409069910a5a2c9462484ba5bfee8f84480"
180
+ },
181
+ {
182
+ "@type": "cr:FileObject",
183
+ "@id": "file-train-00002-of-00018.parquet",
184
+ "name": "train-00002-of-00018.parquet",
185
+ "description": "Published train split shard.",
186
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00002-of-00018.parquet",
187
+ "encodingFormat": "application/x-parquet",
188
+ "contentSize": "2456863463",
189
+ "sha256": "464688c598d77feee7bad1177f6e7c435c4160f3bc786841cdfbcccacb919586"
190
+ },
191
+ {
192
+ "@type": "cr:FileObject",
193
+ "@id": "file-train-00003-of-00018.parquet",
194
+ "name": "train-00003-of-00018.parquet",
195
+ "description": "Published train split shard.",
196
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00003-of-00018.parquet",
197
+ "encodingFormat": "application/x-parquet",
198
+ "contentSize": "2587855930",
199
+ "sha256": "a00f8b77f1ee3063457415bafa820d06fc4cde9a798b38288f0012bfda6094d1"
200
+ },
201
+ {
202
+ "@type": "cr:FileObject",
203
+ "@id": "file-train-00004-of-00018.parquet",
204
+ "name": "train-00004-of-00018.parquet",
205
+ "description": "Published train split shard.",
206
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00004-of-00018.parquet",
207
+ "encodingFormat": "application/x-parquet",
208
+ "contentSize": "2426455080",
209
+ "sha256": "2b70d31abc3454e8024adf4bf68cbd9fe994041326a20727042aa0ed70572014"
210
+ },
211
+ {
212
+ "@type": "cr:FileObject",
213
+ "@id": "file-train-00005-of-00018.parquet",
214
+ "name": "train-00005-of-00018.parquet",
215
+ "description": "Published train split shard.",
216
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00005-of-00018.parquet",
217
+ "encodingFormat": "application/x-parquet",
218
+ "contentSize": "2451685519",
219
+ "sha256": "1bf6168a30b72821bd5ca7d93c9348a777332b243c05bf5a2cbee2b69fe81d54"
220
+ },
221
+ {
222
+ "@type": "cr:FileObject",
223
+ "@id": "file-train-00006-of-00018.parquet",
224
+ "name": "train-00006-of-00018.parquet",
225
+ "description": "Published train split shard.",
226
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00006-of-00018.parquet",
227
+ "encodingFormat": "application/x-parquet",
228
+ "contentSize": "2569546648",
229
+ "sha256": "36cc7c257fcd0e8cde31b4669a413ef16f7ab972180bfa231408d15ba100fa19"
230
+ },
231
+ {
232
+ "@type": "cr:FileObject",
233
+ "@id": "file-train-00007-of-00018.parquet",
234
+ "name": "train-00007-of-00018.parquet",
235
+ "description": "Published train split shard.",
236
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00007-of-00018.parquet",
237
+ "encodingFormat": "application/x-parquet",
238
+ "contentSize": "2352079191",
239
+ "sha256": "b0f878e8ef23c75812c32085deb202b0f05c486f5fd7b5ebbfd2a8751b723b9d"
240
+ },
241
+ {
242
+ "@type": "cr:FileObject",
243
+ "@id": "file-train-00008-of-00018.parquet",
244
+ "name": "train-00008-of-00018.parquet",
245
+ "description": "Published train split shard.",
246
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00008-of-00018.parquet",
247
+ "encodingFormat": "application/x-parquet",
248
+ "contentSize": "2670909898",
249
+ "sha256": "e7632740faac3f6fd806c0bb5b3b1e6642e2fadc152c22685a9ef35f5886d88a"
250
+ },
251
+ {
252
+ "@type": "cr:FileObject",
253
+ "@id": "file-train-00009-of-00018.parquet",
254
+ "name": "train-00009-of-00018.parquet",
255
+ "description": "Published train split shard.",
256
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00009-of-00018.parquet",
257
+ "encodingFormat": "application/x-parquet",
258
+ "contentSize": "2802453098",
259
+ "sha256": "2ed09a8eb21826b30ebe5e7fa8f9705652780239be6be7f84d336ba0a11fde04"
260
+ },
261
+ {
262
+ "@type": "cr:FileObject",
263
+ "@id": "file-train-00010-of-00018.parquet",
264
+ "name": "train-00010-of-00018.parquet",
265
+ "description": "Published train split shard.",
266
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00010-of-00018.parquet",
267
+ "encodingFormat": "application/x-parquet",
268
+ "contentSize": "2692008184",
269
+ "sha256": "2a369a9f0568aa9f4d9271e076c302c93fd65dbcd72e36a76225b5374108f986"
270
+ },
271
+ {
272
+ "@type": "cr:FileObject",
273
+ "@id": "file-train-00011-of-00018.parquet",
274
+ "name": "train-00011-of-00018.parquet",
275
+ "description": "Published train split shard.",
276
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00011-of-00018.parquet",
277
+ "encodingFormat": "application/x-parquet",
278
+ "contentSize": "2789328885",
279
+ "sha256": "91e56a08d866c4b7c2b8258358015ccd7e8dd2ec6554f420dd4cb5b2c2d968da"
280
+ },
281
+ {
282
+ "@type": "cr:FileObject",
283
+ "@id": "file-train-00012-of-00018.parquet",
284
+ "name": "train-00012-of-00018.parquet",
285
+ "description": "Published train split shard.",
286
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00012-of-00018.parquet",
287
+ "encodingFormat": "application/x-parquet",
288
+ "contentSize": "2426247269",
289
+ "sha256": "b4dd54009794a9b47500572aaf8b5a62b61986e954f32fc4f88036f474a401a2"
290
+ },
291
+ {
292
+ "@type": "cr:FileObject",
293
+ "@id": "file-train-00013-of-00018.parquet",
294
+ "name": "train-00013-of-00018.parquet",
295
+ "description": "Published train split shard.",
296
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00013-of-00018.parquet",
297
+ "encodingFormat": "application/x-parquet",
298
+ "contentSize": "2514738691",
299
+ "sha256": "97f227891b111d39692122754b56b95fd7a44fd8150c56502d6672201d1ea0e8"
300
+ },
301
+ {
302
+ "@type": "cr:FileObject",
303
+ "@id": "file-train-00014-of-00018.parquet",
304
+ "name": "train-00014-of-00018.parquet",
305
+ "description": "Published train split shard.",
306
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00014-of-00018.parquet",
307
+ "encodingFormat": "application/x-parquet",
308
+ "contentSize": "2741484482",
309
+ "sha256": "c88624b152e771d7d72698b306e90bdd6545fbd78e702b1b03b52e5980e6fe3a"
310
+ },
311
+ {
312
+ "@type": "cr:FileObject",
313
+ "@id": "file-train-00015-of-00018.parquet",
314
+ "name": "train-00015-of-00018.parquet",
315
+ "description": "Published train split shard.",
316
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00015-of-00018.parquet",
317
+ "encodingFormat": "application/x-parquet",
318
+ "contentSize": "2662963705",
319
+ "sha256": "9f5d83afb66ebe3eea20e33c547b0b5ca53297c7110aa5b6d512cc305855adaa"
320
+ },
321
+ {
322
+ "@type": "cr:FileObject",
323
+ "@id": "file-train-00016-of-00018.parquet",
324
+ "name": "train-00016-of-00018.parquet",
325
+ "description": "Published train split shard.",
326
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00016-of-00018.parquet",
327
+ "encodingFormat": "application/x-parquet",
328
+ "contentSize": "2382833481",
329
+ "sha256": "28209af3d3e43ff34cd3192f6c72287b36997c0225ec9563969f28d028cb503c"
330
+ },
331
+ {
332
+ "@type": "cr:FileObject",
333
+ "@id": "file-train-00017-of-00018.parquet",
334
+ "name": "train-00017-of-00018.parquet",
335
+ "description": "Published train split shard.",
336
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/train/train-00017-of-00018.parquet",
337
+ "encodingFormat": "application/x-parquet",
338
+ "contentSize": "2354622098",
339
+ "sha256": "6ca82b0edbff869e6c5e37c29e11375d37e8a807e0da6345cec1e1f74b003d82"
340
+ },
341
+ {
342
+ "@type": "cr:FileObject",
343
+ "@id": "file-validation-00000-of-00001.parquet",
344
+ "name": "validation-00000-of-00001.parquet",
345
+ "description": "Published validation split shard.",
346
+ "contentUrl": "https://huggingface.co/datasets/RenxingIntelligence/OpenVTON/resolve/afe4443ab7961cabe00ec3122809c89111039cde/validation/validation-00000-of-00001.parquet",
347
+ "encodingFormat": "application/x-parquet",
348
+ "contentSize": "2527219877",
349
+ "sha256": "69f097c36a091dba53f46f845ebf72d6c38bdf02822e70790a21a6f4813e33f4"
350
+ }
351
+ ],
352
+ "recordSet": [
353
+ {
354
+ "@type": "cr:RecordSet",
355
+ "@id": "test-00000-of-00002",
356
+ "name": "test-00000-of-00002",
357
+ "description": "test split, physical shard test/test-00000-of-00002.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
358
+ "field": [
359
+ {
360
+ "@type": "cr:Field",
361
+ "@id": "test-00000-of-00002/source",
362
+ "name": "source",
363
+ "description": "Reference garment image.",
364
+ "dataType": "sc:ImageObject",
365
+ "source": {
366
+ "fileObject": {
367
+ "@id": "file-test-00000-of-00002.parquet"
368
+ },
369
+ "extract": {
370
+ "column": "source"
371
+ },
372
+ "transform": {
373
+ "jsonPath": "bytes"
374
+ }
375
+ }
376
+ },
377
+ {
378
+ "@type": "cr:Field",
379
+ "@id": "test-00000-of-00002/mask",
380
+ "name": "mask",
381
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
382
+ "dataType": "sc:ImageObject",
383
+ "source": {
384
+ "fileObject": {
385
+ "@id": "file-test-00000-of-00002.parquet"
386
+ },
387
+ "extract": {
388
+ "column": "mask"
389
+ },
390
+ "transform": {
391
+ "jsonPath": "bytes"
392
+ }
393
+ }
394
+ },
395
+ {
396
+ "@type": "cr:Field",
397
+ "@id": "test-00000-of-00002/target",
398
+ "name": "target",
399
+ "description": "Person wearing the reference garment (target image).",
400
+ "dataType": "sc:ImageObject",
401
+ "source": {
402
+ "fileObject": {
403
+ "@id": "file-test-00000-of-00002.parquet"
404
+ },
405
+ "extract": {
406
+ "column": "target"
407
+ },
408
+ "transform": {
409
+ "jsonPath": "bytes"
410
+ }
411
+ }
412
+ },
413
+ {
414
+ "@type": "cr:Field",
415
+ "@id": "test-00000-of-00002/caption",
416
+ "name": "caption",
417
+ "description": "Garment description generated with Gemini 2.0 Flash.",
418
+ "dataType": "sc:Text",
419
+ "source": {
420
+ "fileObject": {
421
+ "@id": "file-test-00000-of-00002.parquet"
422
+ },
423
+ "extract": {
424
+ "column": "caption"
425
+ }
426
+ }
427
+ },
428
+ {
429
+ "@type": "cr:Field",
430
+ "@id": "test-00000-of-00002/category",
431
+ "name": "category",
432
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
433
+ "dataType": "sc:Text",
434
+ "source": {
435
+ "fileObject": {
436
+ "@id": "file-test-00000-of-00002.parquet"
437
+ },
438
+ "extract": {
439
+ "column": "category"
440
+ }
441
+ }
442
+ }
443
+ ]
444
+ },
445
+ {
446
+ "@type": "cr:RecordSet",
447
+ "@id": "test-00001-of-00002",
448
+ "name": "test-00001-of-00002",
449
+ "description": "test split, physical shard test/test-00001-of-00002.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
450
+ "field": [
451
+ {
452
+ "@type": "cr:Field",
453
+ "@id": "test-00001-of-00002/source",
454
+ "name": "source",
455
+ "description": "Reference garment image.",
456
+ "dataType": "sc:ImageObject",
457
+ "source": {
458
+ "fileObject": {
459
+ "@id": "file-test-00001-of-00002.parquet"
460
+ },
461
+ "extract": {
462
+ "column": "source"
463
+ },
464
+ "transform": {
465
+ "jsonPath": "bytes"
466
+ }
467
+ }
468
+ },
469
+ {
470
+ "@type": "cr:Field",
471
+ "@id": "test-00001-of-00002/mask",
472
+ "name": "mask",
473
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
474
+ "dataType": "sc:ImageObject",
475
+ "source": {
476
+ "fileObject": {
477
+ "@id": "file-test-00001-of-00002.parquet"
478
+ },
479
+ "extract": {
480
+ "column": "mask"
481
+ },
482
+ "transform": {
483
+ "jsonPath": "bytes"
484
+ }
485
+ }
486
+ },
487
+ {
488
+ "@type": "cr:Field",
489
+ "@id": "test-00001-of-00002/target",
490
+ "name": "target",
491
+ "description": "Person wearing the reference garment (target image).",
492
+ "dataType": "sc:ImageObject",
493
+ "source": {
494
+ "fileObject": {
495
+ "@id": "file-test-00001-of-00002.parquet"
496
+ },
497
+ "extract": {
498
+ "column": "target"
499
+ },
500
+ "transform": {
501
+ "jsonPath": "bytes"
502
+ }
503
+ }
504
+ },
505
+ {
506
+ "@type": "cr:Field",
507
+ "@id": "test-00001-of-00002/caption",
508
+ "name": "caption",
509
+ "description": "Garment description generated with Gemini 2.0 Flash.",
510
+ "dataType": "sc:Text",
511
+ "source": {
512
+ "fileObject": {
513
+ "@id": "file-test-00001-of-00002.parquet"
514
+ },
515
+ "extract": {
516
+ "column": "caption"
517
+ }
518
+ }
519
+ },
520
+ {
521
+ "@type": "cr:Field",
522
+ "@id": "test-00001-of-00002/category",
523
+ "name": "category",
524
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
525
+ "dataType": "sc:Text",
526
+ "source": {
527
+ "fileObject": {
528
+ "@id": "file-test-00001-of-00002.parquet"
529
+ },
530
+ "extract": {
531
+ "column": "category"
532
+ }
533
+ }
534
+ }
535
+ ]
536
+ },
537
+ {
538
+ "@type": "cr:RecordSet",
539
+ "@id": "train-00000-of-00018",
540
+ "name": "train-00000-of-00018",
541
+ "description": "train split, physical shard train/train-00000-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
542
+ "field": [
543
+ {
544
+ "@type": "cr:Field",
545
+ "@id": "train-00000-of-00018/source",
546
+ "name": "source",
547
+ "description": "Reference garment image.",
548
+ "dataType": "sc:ImageObject",
549
+ "source": {
550
+ "fileObject": {
551
+ "@id": "file-train-00000-of-00018.parquet"
552
+ },
553
+ "extract": {
554
+ "column": "source"
555
+ },
556
+ "transform": {
557
+ "jsonPath": "bytes"
558
+ }
559
+ }
560
+ },
561
+ {
562
+ "@type": "cr:Field",
563
+ "@id": "train-00000-of-00018/mask",
564
+ "name": "mask",
565
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
566
+ "dataType": "sc:ImageObject",
567
+ "source": {
568
+ "fileObject": {
569
+ "@id": "file-train-00000-of-00018.parquet"
570
+ },
571
+ "extract": {
572
+ "column": "mask"
573
+ },
574
+ "transform": {
575
+ "jsonPath": "bytes"
576
+ }
577
+ }
578
+ },
579
+ {
580
+ "@type": "cr:Field",
581
+ "@id": "train-00000-of-00018/target",
582
+ "name": "target",
583
+ "description": "Person wearing the reference garment (target image).",
584
+ "dataType": "sc:ImageObject",
585
+ "source": {
586
+ "fileObject": {
587
+ "@id": "file-train-00000-of-00018.parquet"
588
+ },
589
+ "extract": {
590
+ "column": "target"
591
+ },
592
+ "transform": {
593
+ "jsonPath": "bytes"
594
+ }
595
+ }
596
+ },
597
+ {
598
+ "@type": "cr:Field",
599
+ "@id": "train-00000-of-00018/caption",
600
+ "name": "caption",
601
+ "description": "Garment description generated with Gemini 2.0 Flash.",
602
+ "dataType": "sc:Text",
603
+ "source": {
604
+ "fileObject": {
605
+ "@id": "file-train-00000-of-00018.parquet"
606
+ },
607
+ "extract": {
608
+ "column": "caption"
609
+ }
610
+ }
611
+ },
612
+ {
613
+ "@type": "cr:Field",
614
+ "@id": "train-00000-of-00018/category",
615
+ "name": "category",
616
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
617
+ "dataType": "sc:Text",
618
+ "source": {
619
+ "fileObject": {
620
+ "@id": "file-train-00000-of-00018.parquet"
621
+ },
622
+ "extract": {
623
+ "column": "category"
624
+ }
625
+ }
626
+ }
627
+ ]
628
+ },
629
+ {
630
+ "@type": "cr:RecordSet",
631
+ "@id": "train-00001-of-00018",
632
+ "name": "train-00001-of-00018",
633
+ "description": "train split, physical shard train/train-00001-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
634
+ "field": [
635
+ {
636
+ "@type": "cr:Field",
637
+ "@id": "train-00001-of-00018/source",
638
+ "name": "source",
639
+ "description": "Reference garment image.",
640
+ "dataType": "sc:ImageObject",
641
+ "source": {
642
+ "fileObject": {
643
+ "@id": "file-train-00001-of-00018.parquet"
644
+ },
645
+ "extract": {
646
+ "column": "source"
647
+ },
648
+ "transform": {
649
+ "jsonPath": "bytes"
650
+ }
651
+ }
652
+ },
653
+ {
654
+ "@type": "cr:Field",
655
+ "@id": "train-00001-of-00018/mask",
656
+ "name": "mask",
657
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
658
+ "dataType": "sc:ImageObject",
659
+ "source": {
660
+ "fileObject": {
661
+ "@id": "file-train-00001-of-00018.parquet"
662
+ },
663
+ "extract": {
664
+ "column": "mask"
665
+ },
666
+ "transform": {
667
+ "jsonPath": "bytes"
668
+ }
669
+ }
670
+ },
671
+ {
672
+ "@type": "cr:Field",
673
+ "@id": "train-00001-of-00018/target",
674
+ "name": "target",
675
+ "description": "Person wearing the reference garment (target image).",
676
+ "dataType": "sc:ImageObject",
677
+ "source": {
678
+ "fileObject": {
679
+ "@id": "file-train-00001-of-00018.parquet"
680
+ },
681
+ "extract": {
682
+ "column": "target"
683
+ },
684
+ "transform": {
685
+ "jsonPath": "bytes"
686
+ }
687
+ }
688
+ },
689
+ {
690
+ "@type": "cr:Field",
691
+ "@id": "train-00001-of-00018/caption",
692
+ "name": "caption",
693
+ "description": "Garment description generated with Gemini 2.0 Flash.",
694
+ "dataType": "sc:Text",
695
+ "source": {
696
+ "fileObject": {
697
+ "@id": "file-train-00001-of-00018.parquet"
698
+ },
699
+ "extract": {
700
+ "column": "caption"
701
+ }
702
+ }
703
+ },
704
+ {
705
+ "@type": "cr:Field",
706
+ "@id": "train-00001-of-00018/category",
707
+ "name": "category",
708
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
709
+ "dataType": "sc:Text",
710
+ "source": {
711
+ "fileObject": {
712
+ "@id": "file-train-00001-of-00018.parquet"
713
+ },
714
+ "extract": {
715
+ "column": "category"
716
+ }
717
+ }
718
+ }
719
+ ]
720
+ },
721
+ {
722
+ "@type": "cr:RecordSet",
723
+ "@id": "train-00002-of-00018",
724
+ "name": "train-00002-of-00018",
725
+ "description": "train split, physical shard train/train-00002-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
726
+ "field": [
727
+ {
728
+ "@type": "cr:Field",
729
+ "@id": "train-00002-of-00018/source",
730
+ "name": "source",
731
+ "description": "Reference garment image.",
732
+ "dataType": "sc:ImageObject",
733
+ "source": {
734
+ "fileObject": {
735
+ "@id": "file-train-00002-of-00018.parquet"
736
+ },
737
+ "extract": {
738
+ "column": "source"
739
+ },
740
+ "transform": {
741
+ "jsonPath": "bytes"
742
+ }
743
+ }
744
+ },
745
+ {
746
+ "@type": "cr:Field",
747
+ "@id": "train-00002-of-00018/mask",
748
+ "name": "mask",
749
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
750
+ "dataType": "sc:ImageObject",
751
+ "source": {
752
+ "fileObject": {
753
+ "@id": "file-train-00002-of-00018.parquet"
754
+ },
755
+ "extract": {
756
+ "column": "mask"
757
+ },
758
+ "transform": {
759
+ "jsonPath": "bytes"
760
+ }
761
+ }
762
+ },
763
+ {
764
+ "@type": "cr:Field",
765
+ "@id": "train-00002-of-00018/target",
766
+ "name": "target",
767
+ "description": "Person wearing the reference garment (target image).",
768
+ "dataType": "sc:ImageObject",
769
+ "source": {
770
+ "fileObject": {
771
+ "@id": "file-train-00002-of-00018.parquet"
772
+ },
773
+ "extract": {
774
+ "column": "target"
775
+ },
776
+ "transform": {
777
+ "jsonPath": "bytes"
778
+ }
779
+ }
780
+ },
781
+ {
782
+ "@type": "cr:Field",
783
+ "@id": "train-00002-of-00018/caption",
784
+ "name": "caption",
785
+ "description": "Garment description generated with Gemini 2.0 Flash.",
786
+ "dataType": "sc:Text",
787
+ "source": {
788
+ "fileObject": {
789
+ "@id": "file-train-00002-of-00018.parquet"
790
+ },
791
+ "extract": {
792
+ "column": "caption"
793
+ }
794
+ }
795
+ },
796
+ {
797
+ "@type": "cr:Field",
798
+ "@id": "train-00002-of-00018/category",
799
+ "name": "category",
800
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
801
+ "dataType": "sc:Text",
802
+ "source": {
803
+ "fileObject": {
804
+ "@id": "file-train-00002-of-00018.parquet"
805
+ },
806
+ "extract": {
807
+ "column": "category"
808
+ }
809
+ }
810
+ }
811
+ ]
812
+ },
813
+ {
814
+ "@type": "cr:RecordSet",
815
+ "@id": "train-00003-of-00018",
816
+ "name": "train-00003-of-00018",
817
+ "description": "train split, physical shard train/train-00003-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
818
+ "field": [
819
+ {
820
+ "@type": "cr:Field",
821
+ "@id": "train-00003-of-00018/source",
822
+ "name": "source",
823
+ "description": "Reference garment image.",
824
+ "dataType": "sc:ImageObject",
825
+ "source": {
826
+ "fileObject": {
827
+ "@id": "file-train-00003-of-00018.parquet"
828
+ },
829
+ "extract": {
830
+ "column": "source"
831
+ },
832
+ "transform": {
833
+ "jsonPath": "bytes"
834
+ }
835
+ }
836
+ },
837
+ {
838
+ "@type": "cr:Field",
839
+ "@id": "train-00003-of-00018/mask",
840
+ "name": "mask",
841
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
842
+ "dataType": "sc:ImageObject",
843
+ "source": {
844
+ "fileObject": {
845
+ "@id": "file-train-00003-of-00018.parquet"
846
+ },
847
+ "extract": {
848
+ "column": "mask"
849
+ },
850
+ "transform": {
851
+ "jsonPath": "bytes"
852
+ }
853
+ }
854
+ },
855
+ {
856
+ "@type": "cr:Field",
857
+ "@id": "train-00003-of-00018/target",
858
+ "name": "target",
859
+ "description": "Person wearing the reference garment (target image).",
860
+ "dataType": "sc:ImageObject",
861
+ "source": {
862
+ "fileObject": {
863
+ "@id": "file-train-00003-of-00018.parquet"
864
+ },
865
+ "extract": {
866
+ "column": "target"
867
+ },
868
+ "transform": {
869
+ "jsonPath": "bytes"
870
+ }
871
+ }
872
+ },
873
+ {
874
+ "@type": "cr:Field",
875
+ "@id": "train-00003-of-00018/caption",
876
+ "name": "caption",
877
+ "description": "Garment description generated with Gemini 2.0 Flash.",
878
+ "dataType": "sc:Text",
879
+ "source": {
880
+ "fileObject": {
881
+ "@id": "file-train-00003-of-00018.parquet"
882
+ },
883
+ "extract": {
884
+ "column": "caption"
885
+ }
886
+ }
887
+ },
888
+ {
889
+ "@type": "cr:Field",
890
+ "@id": "train-00003-of-00018/category",
891
+ "name": "category",
892
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
893
+ "dataType": "sc:Text",
894
+ "source": {
895
+ "fileObject": {
896
+ "@id": "file-train-00003-of-00018.parquet"
897
+ },
898
+ "extract": {
899
+ "column": "category"
900
+ }
901
+ }
902
+ }
903
+ ]
904
+ },
905
+ {
906
+ "@type": "cr:RecordSet",
907
+ "@id": "train-00004-of-00018",
908
+ "name": "train-00004-of-00018",
909
+ "description": "train split, physical shard train/train-00004-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
910
+ "field": [
911
+ {
912
+ "@type": "cr:Field",
913
+ "@id": "train-00004-of-00018/source",
914
+ "name": "source",
915
+ "description": "Reference garment image.",
916
+ "dataType": "sc:ImageObject",
917
+ "source": {
918
+ "fileObject": {
919
+ "@id": "file-train-00004-of-00018.parquet"
920
+ },
921
+ "extract": {
922
+ "column": "source"
923
+ },
924
+ "transform": {
925
+ "jsonPath": "bytes"
926
+ }
927
+ }
928
+ },
929
+ {
930
+ "@type": "cr:Field",
931
+ "@id": "train-00004-of-00018/mask",
932
+ "name": "mask",
933
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
934
+ "dataType": "sc:ImageObject",
935
+ "source": {
936
+ "fileObject": {
937
+ "@id": "file-train-00004-of-00018.parquet"
938
+ },
939
+ "extract": {
940
+ "column": "mask"
941
+ },
942
+ "transform": {
943
+ "jsonPath": "bytes"
944
+ }
945
+ }
946
+ },
947
+ {
948
+ "@type": "cr:Field",
949
+ "@id": "train-00004-of-00018/target",
950
+ "name": "target",
951
+ "description": "Person wearing the reference garment (target image).",
952
+ "dataType": "sc:ImageObject",
953
+ "source": {
954
+ "fileObject": {
955
+ "@id": "file-train-00004-of-00018.parquet"
956
+ },
957
+ "extract": {
958
+ "column": "target"
959
+ },
960
+ "transform": {
961
+ "jsonPath": "bytes"
962
+ }
963
+ }
964
+ },
965
+ {
966
+ "@type": "cr:Field",
967
+ "@id": "train-00004-of-00018/caption",
968
+ "name": "caption",
969
+ "description": "Garment description generated with Gemini 2.0 Flash.",
970
+ "dataType": "sc:Text",
971
+ "source": {
972
+ "fileObject": {
973
+ "@id": "file-train-00004-of-00018.parquet"
974
+ },
975
+ "extract": {
976
+ "column": "caption"
977
+ }
978
+ }
979
+ },
980
+ {
981
+ "@type": "cr:Field",
982
+ "@id": "train-00004-of-00018/category",
983
+ "name": "category",
984
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
985
+ "dataType": "sc:Text",
986
+ "source": {
987
+ "fileObject": {
988
+ "@id": "file-train-00004-of-00018.parquet"
989
+ },
990
+ "extract": {
991
+ "column": "category"
992
+ }
993
+ }
994
+ }
995
+ ]
996
+ },
997
+ {
998
+ "@type": "cr:RecordSet",
999
+ "@id": "train-00005-of-00018",
1000
+ "name": "train-00005-of-00018",
1001
+ "description": "train split, physical shard train/train-00005-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1002
+ "field": [
1003
+ {
1004
+ "@type": "cr:Field",
1005
+ "@id": "train-00005-of-00018/source",
1006
+ "name": "source",
1007
+ "description": "Reference garment image.",
1008
+ "dataType": "sc:ImageObject",
1009
+ "source": {
1010
+ "fileObject": {
1011
+ "@id": "file-train-00005-of-00018.parquet"
1012
+ },
1013
+ "extract": {
1014
+ "column": "source"
1015
+ },
1016
+ "transform": {
1017
+ "jsonPath": "bytes"
1018
+ }
1019
+ }
1020
+ },
1021
+ {
1022
+ "@type": "cr:Field",
1023
+ "@id": "train-00005-of-00018/mask",
1024
+ "name": "mask",
1025
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1026
+ "dataType": "sc:ImageObject",
1027
+ "source": {
1028
+ "fileObject": {
1029
+ "@id": "file-train-00005-of-00018.parquet"
1030
+ },
1031
+ "extract": {
1032
+ "column": "mask"
1033
+ },
1034
+ "transform": {
1035
+ "jsonPath": "bytes"
1036
+ }
1037
+ }
1038
+ },
1039
+ {
1040
+ "@type": "cr:Field",
1041
+ "@id": "train-00005-of-00018/target",
1042
+ "name": "target",
1043
+ "description": "Person wearing the reference garment (target image).",
1044
+ "dataType": "sc:ImageObject",
1045
+ "source": {
1046
+ "fileObject": {
1047
+ "@id": "file-train-00005-of-00018.parquet"
1048
+ },
1049
+ "extract": {
1050
+ "column": "target"
1051
+ },
1052
+ "transform": {
1053
+ "jsonPath": "bytes"
1054
+ }
1055
+ }
1056
+ },
1057
+ {
1058
+ "@type": "cr:Field",
1059
+ "@id": "train-00005-of-00018/caption",
1060
+ "name": "caption",
1061
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1062
+ "dataType": "sc:Text",
1063
+ "source": {
1064
+ "fileObject": {
1065
+ "@id": "file-train-00005-of-00018.parquet"
1066
+ },
1067
+ "extract": {
1068
+ "column": "caption"
1069
+ }
1070
+ }
1071
+ },
1072
+ {
1073
+ "@type": "cr:Field",
1074
+ "@id": "train-00005-of-00018/category",
1075
+ "name": "category",
1076
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1077
+ "dataType": "sc:Text",
1078
+ "source": {
1079
+ "fileObject": {
1080
+ "@id": "file-train-00005-of-00018.parquet"
1081
+ },
1082
+ "extract": {
1083
+ "column": "category"
1084
+ }
1085
+ }
1086
+ }
1087
+ ]
1088
+ },
1089
+ {
1090
+ "@type": "cr:RecordSet",
1091
+ "@id": "train-00006-of-00018",
1092
+ "name": "train-00006-of-00018",
1093
+ "description": "train split, physical shard train/train-00006-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1094
+ "field": [
1095
+ {
1096
+ "@type": "cr:Field",
1097
+ "@id": "train-00006-of-00018/source",
1098
+ "name": "source",
1099
+ "description": "Reference garment image.",
1100
+ "dataType": "sc:ImageObject",
1101
+ "source": {
1102
+ "fileObject": {
1103
+ "@id": "file-train-00006-of-00018.parquet"
1104
+ },
1105
+ "extract": {
1106
+ "column": "source"
1107
+ },
1108
+ "transform": {
1109
+ "jsonPath": "bytes"
1110
+ }
1111
+ }
1112
+ },
1113
+ {
1114
+ "@type": "cr:Field",
1115
+ "@id": "train-00006-of-00018/mask",
1116
+ "name": "mask",
1117
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1118
+ "dataType": "sc:ImageObject",
1119
+ "source": {
1120
+ "fileObject": {
1121
+ "@id": "file-train-00006-of-00018.parquet"
1122
+ },
1123
+ "extract": {
1124
+ "column": "mask"
1125
+ },
1126
+ "transform": {
1127
+ "jsonPath": "bytes"
1128
+ }
1129
+ }
1130
+ },
1131
+ {
1132
+ "@type": "cr:Field",
1133
+ "@id": "train-00006-of-00018/target",
1134
+ "name": "target",
1135
+ "description": "Person wearing the reference garment (target image).",
1136
+ "dataType": "sc:ImageObject",
1137
+ "source": {
1138
+ "fileObject": {
1139
+ "@id": "file-train-00006-of-00018.parquet"
1140
+ },
1141
+ "extract": {
1142
+ "column": "target"
1143
+ },
1144
+ "transform": {
1145
+ "jsonPath": "bytes"
1146
+ }
1147
+ }
1148
+ },
1149
+ {
1150
+ "@type": "cr:Field",
1151
+ "@id": "train-00006-of-00018/caption",
1152
+ "name": "caption",
1153
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1154
+ "dataType": "sc:Text",
1155
+ "source": {
1156
+ "fileObject": {
1157
+ "@id": "file-train-00006-of-00018.parquet"
1158
+ },
1159
+ "extract": {
1160
+ "column": "caption"
1161
+ }
1162
+ }
1163
+ },
1164
+ {
1165
+ "@type": "cr:Field",
1166
+ "@id": "train-00006-of-00018/category",
1167
+ "name": "category",
1168
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1169
+ "dataType": "sc:Text",
1170
+ "source": {
1171
+ "fileObject": {
1172
+ "@id": "file-train-00006-of-00018.parquet"
1173
+ },
1174
+ "extract": {
1175
+ "column": "category"
1176
+ }
1177
+ }
1178
+ }
1179
+ ]
1180
+ },
1181
+ {
1182
+ "@type": "cr:RecordSet",
1183
+ "@id": "train-00007-of-00018",
1184
+ "name": "train-00007-of-00018",
1185
+ "description": "train split, physical shard train/train-00007-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1186
+ "field": [
1187
+ {
1188
+ "@type": "cr:Field",
1189
+ "@id": "train-00007-of-00018/source",
1190
+ "name": "source",
1191
+ "description": "Reference garment image.",
1192
+ "dataType": "sc:ImageObject",
1193
+ "source": {
1194
+ "fileObject": {
1195
+ "@id": "file-train-00007-of-00018.parquet"
1196
+ },
1197
+ "extract": {
1198
+ "column": "source"
1199
+ },
1200
+ "transform": {
1201
+ "jsonPath": "bytes"
1202
+ }
1203
+ }
1204
+ },
1205
+ {
1206
+ "@type": "cr:Field",
1207
+ "@id": "train-00007-of-00018/mask",
1208
+ "name": "mask",
1209
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1210
+ "dataType": "sc:ImageObject",
1211
+ "source": {
1212
+ "fileObject": {
1213
+ "@id": "file-train-00007-of-00018.parquet"
1214
+ },
1215
+ "extract": {
1216
+ "column": "mask"
1217
+ },
1218
+ "transform": {
1219
+ "jsonPath": "bytes"
1220
+ }
1221
+ }
1222
+ },
1223
+ {
1224
+ "@type": "cr:Field",
1225
+ "@id": "train-00007-of-00018/target",
1226
+ "name": "target",
1227
+ "description": "Person wearing the reference garment (target image).",
1228
+ "dataType": "sc:ImageObject",
1229
+ "source": {
1230
+ "fileObject": {
1231
+ "@id": "file-train-00007-of-00018.parquet"
1232
+ },
1233
+ "extract": {
1234
+ "column": "target"
1235
+ },
1236
+ "transform": {
1237
+ "jsonPath": "bytes"
1238
+ }
1239
+ }
1240
+ },
1241
+ {
1242
+ "@type": "cr:Field",
1243
+ "@id": "train-00007-of-00018/caption",
1244
+ "name": "caption",
1245
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1246
+ "dataType": "sc:Text",
1247
+ "source": {
1248
+ "fileObject": {
1249
+ "@id": "file-train-00007-of-00018.parquet"
1250
+ },
1251
+ "extract": {
1252
+ "column": "caption"
1253
+ }
1254
+ }
1255
+ },
1256
+ {
1257
+ "@type": "cr:Field",
1258
+ "@id": "train-00007-of-00018/category",
1259
+ "name": "category",
1260
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1261
+ "dataType": "sc:Text",
1262
+ "source": {
1263
+ "fileObject": {
1264
+ "@id": "file-train-00007-of-00018.parquet"
1265
+ },
1266
+ "extract": {
1267
+ "column": "category"
1268
+ }
1269
+ }
1270
+ }
1271
+ ]
1272
+ },
1273
+ {
1274
+ "@type": "cr:RecordSet",
1275
+ "@id": "train-00008-of-00018",
1276
+ "name": "train-00008-of-00018",
1277
+ "description": "train split, physical shard train/train-00008-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1278
+ "field": [
1279
+ {
1280
+ "@type": "cr:Field",
1281
+ "@id": "train-00008-of-00018/source",
1282
+ "name": "source",
1283
+ "description": "Reference garment image.",
1284
+ "dataType": "sc:ImageObject",
1285
+ "source": {
1286
+ "fileObject": {
1287
+ "@id": "file-train-00008-of-00018.parquet"
1288
+ },
1289
+ "extract": {
1290
+ "column": "source"
1291
+ },
1292
+ "transform": {
1293
+ "jsonPath": "bytes"
1294
+ }
1295
+ }
1296
+ },
1297
+ {
1298
+ "@type": "cr:Field",
1299
+ "@id": "train-00008-of-00018/mask",
1300
+ "name": "mask",
1301
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1302
+ "dataType": "sc:ImageObject",
1303
+ "source": {
1304
+ "fileObject": {
1305
+ "@id": "file-train-00008-of-00018.parquet"
1306
+ },
1307
+ "extract": {
1308
+ "column": "mask"
1309
+ },
1310
+ "transform": {
1311
+ "jsonPath": "bytes"
1312
+ }
1313
+ }
1314
+ },
1315
+ {
1316
+ "@type": "cr:Field",
1317
+ "@id": "train-00008-of-00018/target",
1318
+ "name": "target",
1319
+ "description": "Person wearing the reference garment (target image).",
1320
+ "dataType": "sc:ImageObject",
1321
+ "source": {
1322
+ "fileObject": {
1323
+ "@id": "file-train-00008-of-00018.parquet"
1324
+ },
1325
+ "extract": {
1326
+ "column": "target"
1327
+ },
1328
+ "transform": {
1329
+ "jsonPath": "bytes"
1330
+ }
1331
+ }
1332
+ },
1333
+ {
1334
+ "@type": "cr:Field",
1335
+ "@id": "train-00008-of-00018/caption",
1336
+ "name": "caption",
1337
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1338
+ "dataType": "sc:Text",
1339
+ "source": {
1340
+ "fileObject": {
1341
+ "@id": "file-train-00008-of-00018.parquet"
1342
+ },
1343
+ "extract": {
1344
+ "column": "caption"
1345
+ }
1346
+ }
1347
+ },
1348
+ {
1349
+ "@type": "cr:Field",
1350
+ "@id": "train-00008-of-00018/category",
1351
+ "name": "category",
1352
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1353
+ "dataType": "sc:Text",
1354
+ "source": {
1355
+ "fileObject": {
1356
+ "@id": "file-train-00008-of-00018.parquet"
1357
+ },
1358
+ "extract": {
1359
+ "column": "category"
1360
+ }
1361
+ }
1362
+ }
1363
+ ]
1364
+ },
1365
+ {
1366
+ "@type": "cr:RecordSet",
1367
+ "@id": "train-00009-of-00018",
1368
+ "name": "train-00009-of-00018",
1369
+ "description": "train split, physical shard train/train-00009-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1370
+ "field": [
1371
+ {
1372
+ "@type": "cr:Field",
1373
+ "@id": "train-00009-of-00018/source",
1374
+ "name": "source",
1375
+ "description": "Reference garment image.",
1376
+ "dataType": "sc:ImageObject",
1377
+ "source": {
1378
+ "fileObject": {
1379
+ "@id": "file-train-00009-of-00018.parquet"
1380
+ },
1381
+ "extract": {
1382
+ "column": "source"
1383
+ },
1384
+ "transform": {
1385
+ "jsonPath": "bytes"
1386
+ }
1387
+ }
1388
+ },
1389
+ {
1390
+ "@type": "cr:Field",
1391
+ "@id": "train-00009-of-00018/mask",
1392
+ "name": "mask",
1393
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1394
+ "dataType": "sc:ImageObject",
1395
+ "source": {
1396
+ "fileObject": {
1397
+ "@id": "file-train-00009-of-00018.parquet"
1398
+ },
1399
+ "extract": {
1400
+ "column": "mask"
1401
+ },
1402
+ "transform": {
1403
+ "jsonPath": "bytes"
1404
+ }
1405
+ }
1406
+ },
1407
+ {
1408
+ "@type": "cr:Field",
1409
+ "@id": "train-00009-of-00018/target",
1410
+ "name": "target",
1411
+ "description": "Person wearing the reference garment (target image).",
1412
+ "dataType": "sc:ImageObject",
1413
+ "source": {
1414
+ "fileObject": {
1415
+ "@id": "file-train-00009-of-00018.parquet"
1416
+ },
1417
+ "extract": {
1418
+ "column": "target"
1419
+ },
1420
+ "transform": {
1421
+ "jsonPath": "bytes"
1422
+ }
1423
+ }
1424
+ },
1425
+ {
1426
+ "@type": "cr:Field",
1427
+ "@id": "train-00009-of-00018/caption",
1428
+ "name": "caption",
1429
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1430
+ "dataType": "sc:Text",
1431
+ "source": {
1432
+ "fileObject": {
1433
+ "@id": "file-train-00009-of-00018.parquet"
1434
+ },
1435
+ "extract": {
1436
+ "column": "caption"
1437
+ }
1438
+ }
1439
+ },
1440
+ {
1441
+ "@type": "cr:Field",
1442
+ "@id": "train-00009-of-00018/category",
1443
+ "name": "category",
1444
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1445
+ "dataType": "sc:Text",
1446
+ "source": {
1447
+ "fileObject": {
1448
+ "@id": "file-train-00009-of-00018.parquet"
1449
+ },
1450
+ "extract": {
1451
+ "column": "category"
1452
+ }
1453
+ }
1454
+ }
1455
+ ]
1456
+ },
1457
+ {
1458
+ "@type": "cr:RecordSet",
1459
+ "@id": "train-00010-of-00018",
1460
+ "name": "train-00010-of-00018",
1461
+ "description": "train split, physical shard train/train-00010-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1462
+ "field": [
1463
+ {
1464
+ "@type": "cr:Field",
1465
+ "@id": "train-00010-of-00018/source",
1466
+ "name": "source",
1467
+ "description": "Reference garment image.",
1468
+ "dataType": "sc:ImageObject",
1469
+ "source": {
1470
+ "fileObject": {
1471
+ "@id": "file-train-00010-of-00018.parquet"
1472
+ },
1473
+ "extract": {
1474
+ "column": "source"
1475
+ },
1476
+ "transform": {
1477
+ "jsonPath": "bytes"
1478
+ }
1479
+ }
1480
+ },
1481
+ {
1482
+ "@type": "cr:Field",
1483
+ "@id": "train-00010-of-00018/mask",
1484
+ "name": "mask",
1485
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1486
+ "dataType": "sc:ImageObject",
1487
+ "source": {
1488
+ "fileObject": {
1489
+ "@id": "file-train-00010-of-00018.parquet"
1490
+ },
1491
+ "extract": {
1492
+ "column": "mask"
1493
+ },
1494
+ "transform": {
1495
+ "jsonPath": "bytes"
1496
+ }
1497
+ }
1498
+ },
1499
+ {
1500
+ "@type": "cr:Field",
1501
+ "@id": "train-00010-of-00018/target",
1502
+ "name": "target",
1503
+ "description": "Person wearing the reference garment (target image).",
1504
+ "dataType": "sc:ImageObject",
1505
+ "source": {
1506
+ "fileObject": {
1507
+ "@id": "file-train-00010-of-00018.parquet"
1508
+ },
1509
+ "extract": {
1510
+ "column": "target"
1511
+ },
1512
+ "transform": {
1513
+ "jsonPath": "bytes"
1514
+ }
1515
+ }
1516
+ },
1517
+ {
1518
+ "@type": "cr:Field",
1519
+ "@id": "train-00010-of-00018/caption",
1520
+ "name": "caption",
1521
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1522
+ "dataType": "sc:Text",
1523
+ "source": {
1524
+ "fileObject": {
1525
+ "@id": "file-train-00010-of-00018.parquet"
1526
+ },
1527
+ "extract": {
1528
+ "column": "caption"
1529
+ }
1530
+ }
1531
+ },
1532
+ {
1533
+ "@type": "cr:Field",
1534
+ "@id": "train-00010-of-00018/category",
1535
+ "name": "category",
1536
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1537
+ "dataType": "sc:Text",
1538
+ "source": {
1539
+ "fileObject": {
1540
+ "@id": "file-train-00010-of-00018.parquet"
1541
+ },
1542
+ "extract": {
1543
+ "column": "category"
1544
+ }
1545
+ }
1546
+ }
1547
+ ]
1548
+ },
1549
+ {
1550
+ "@type": "cr:RecordSet",
1551
+ "@id": "train-00011-of-00018",
1552
+ "name": "train-00011-of-00018",
1553
+ "description": "train split, physical shard train/train-00011-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1554
+ "field": [
1555
+ {
1556
+ "@type": "cr:Field",
1557
+ "@id": "train-00011-of-00018/source",
1558
+ "name": "source",
1559
+ "description": "Reference garment image.",
1560
+ "dataType": "sc:ImageObject",
1561
+ "source": {
1562
+ "fileObject": {
1563
+ "@id": "file-train-00011-of-00018.parquet"
1564
+ },
1565
+ "extract": {
1566
+ "column": "source"
1567
+ },
1568
+ "transform": {
1569
+ "jsonPath": "bytes"
1570
+ }
1571
+ }
1572
+ },
1573
+ {
1574
+ "@type": "cr:Field",
1575
+ "@id": "train-00011-of-00018/mask",
1576
+ "name": "mask",
1577
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1578
+ "dataType": "sc:ImageObject",
1579
+ "source": {
1580
+ "fileObject": {
1581
+ "@id": "file-train-00011-of-00018.parquet"
1582
+ },
1583
+ "extract": {
1584
+ "column": "mask"
1585
+ },
1586
+ "transform": {
1587
+ "jsonPath": "bytes"
1588
+ }
1589
+ }
1590
+ },
1591
+ {
1592
+ "@type": "cr:Field",
1593
+ "@id": "train-00011-of-00018/target",
1594
+ "name": "target",
1595
+ "description": "Person wearing the reference garment (target image).",
1596
+ "dataType": "sc:ImageObject",
1597
+ "source": {
1598
+ "fileObject": {
1599
+ "@id": "file-train-00011-of-00018.parquet"
1600
+ },
1601
+ "extract": {
1602
+ "column": "target"
1603
+ },
1604
+ "transform": {
1605
+ "jsonPath": "bytes"
1606
+ }
1607
+ }
1608
+ },
1609
+ {
1610
+ "@type": "cr:Field",
1611
+ "@id": "train-00011-of-00018/caption",
1612
+ "name": "caption",
1613
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1614
+ "dataType": "sc:Text",
1615
+ "source": {
1616
+ "fileObject": {
1617
+ "@id": "file-train-00011-of-00018.parquet"
1618
+ },
1619
+ "extract": {
1620
+ "column": "caption"
1621
+ }
1622
+ }
1623
+ },
1624
+ {
1625
+ "@type": "cr:Field",
1626
+ "@id": "train-00011-of-00018/category",
1627
+ "name": "category",
1628
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1629
+ "dataType": "sc:Text",
1630
+ "source": {
1631
+ "fileObject": {
1632
+ "@id": "file-train-00011-of-00018.parquet"
1633
+ },
1634
+ "extract": {
1635
+ "column": "category"
1636
+ }
1637
+ }
1638
+ }
1639
+ ]
1640
+ },
1641
+ {
1642
+ "@type": "cr:RecordSet",
1643
+ "@id": "train-00012-of-00018",
1644
+ "name": "train-00012-of-00018",
1645
+ "description": "train split, physical shard train/train-00012-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1646
+ "field": [
1647
+ {
1648
+ "@type": "cr:Field",
1649
+ "@id": "train-00012-of-00018/source",
1650
+ "name": "source",
1651
+ "description": "Reference garment image.",
1652
+ "dataType": "sc:ImageObject",
1653
+ "source": {
1654
+ "fileObject": {
1655
+ "@id": "file-train-00012-of-00018.parquet"
1656
+ },
1657
+ "extract": {
1658
+ "column": "source"
1659
+ },
1660
+ "transform": {
1661
+ "jsonPath": "bytes"
1662
+ }
1663
+ }
1664
+ },
1665
+ {
1666
+ "@type": "cr:Field",
1667
+ "@id": "train-00012-of-00018/mask",
1668
+ "name": "mask",
1669
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1670
+ "dataType": "sc:ImageObject",
1671
+ "source": {
1672
+ "fileObject": {
1673
+ "@id": "file-train-00012-of-00018.parquet"
1674
+ },
1675
+ "extract": {
1676
+ "column": "mask"
1677
+ },
1678
+ "transform": {
1679
+ "jsonPath": "bytes"
1680
+ }
1681
+ }
1682
+ },
1683
+ {
1684
+ "@type": "cr:Field",
1685
+ "@id": "train-00012-of-00018/target",
1686
+ "name": "target",
1687
+ "description": "Person wearing the reference garment (target image).",
1688
+ "dataType": "sc:ImageObject",
1689
+ "source": {
1690
+ "fileObject": {
1691
+ "@id": "file-train-00012-of-00018.parquet"
1692
+ },
1693
+ "extract": {
1694
+ "column": "target"
1695
+ },
1696
+ "transform": {
1697
+ "jsonPath": "bytes"
1698
+ }
1699
+ }
1700
+ },
1701
+ {
1702
+ "@type": "cr:Field",
1703
+ "@id": "train-00012-of-00018/caption",
1704
+ "name": "caption",
1705
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1706
+ "dataType": "sc:Text",
1707
+ "source": {
1708
+ "fileObject": {
1709
+ "@id": "file-train-00012-of-00018.parquet"
1710
+ },
1711
+ "extract": {
1712
+ "column": "caption"
1713
+ }
1714
+ }
1715
+ },
1716
+ {
1717
+ "@type": "cr:Field",
1718
+ "@id": "train-00012-of-00018/category",
1719
+ "name": "category",
1720
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1721
+ "dataType": "sc:Text",
1722
+ "source": {
1723
+ "fileObject": {
1724
+ "@id": "file-train-00012-of-00018.parquet"
1725
+ },
1726
+ "extract": {
1727
+ "column": "category"
1728
+ }
1729
+ }
1730
+ }
1731
+ ]
1732
+ },
1733
+ {
1734
+ "@type": "cr:RecordSet",
1735
+ "@id": "train-00013-of-00018",
1736
+ "name": "train-00013-of-00018",
1737
+ "description": "train split, physical shard train/train-00013-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1738
+ "field": [
1739
+ {
1740
+ "@type": "cr:Field",
1741
+ "@id": "train-00013-of-00018/source",
1742
+ "name": "source",
1743
+ "description": "Reference garment image.",
1744
+ "dataType": "sc:ImageObject",
1745
+ "source": {
1746
+ "fileObject": {
1747
+ "@id": "file-train-00013-of-00018.parquet"
1748
+ },
1749
+ "extract": {
1750
+ "column": "source"
1751
+ },
1752
+ "transform": {
1753
+ "jsonPath": "bytes"
1754
+ }
1755
+ }
1756
+ },
1757
+ {
1758
+ "@type": "cr:Field",
1759
+ "@id": "train-00013-of-00018/mask",
1760
+ "name": "mask",
1761
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1762
+ "dataType": "sc:ImageObject",
1763
+ "source": {
1764
+ "fileObject": {
1765
+ "@id": "file-train-00013-of-00018.parquet"
1766
+ },
1767
+ "extract": {
1768
+ "column": "mask"
1769
+ },
1770
+ "transform": {
1771
+ "jsonPath": "bytes"
1772
+ }
1773
+ }
1774
+ },
1775
+ {
1776
+ "@type": "cr:Field",
1777
+ "@id": "train-00013-of-00018/target",
1778
+ "name": "target",
1779
+ "description": "Person wearing the reference garment (target image).",
1780
+ "dataType": "sc:ImageObject",
1781
+ "source": {
1782
+ "fileObject": {
1783
+ "@id": "file-train-00013-of-00018.parquet"
1784
+ },
1785
+ "extract": {
1786
+ "column": "target"
1787
+ },
1788
+ "transform": {
1789
+ "jsonPath": "bytes"
1790
+ }
1791
+ }
1792
+ },
1793
+ {
1794
+ "@type": "cr:Field",
1795
+ "@id": "train-00013-of-00018/caption",
1796
+ "name": "caption",
1797
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1798
+ "dataType": "sc:Text",
1799
+ "source": {
1800
+ "fileObject": {
1801
+ "@id": "file-train-00013-of-00018.parquet"
1802
+ },
1803
+ "extract": {
1804
+ "column": "caption"
1805
+ }
1806
+ }
1807
+ },
1808
+ {
1809
+ "@type": "cr:Field",
1810
+ "@id": "train-00013-of-00018/category",
1811
+ "name": "category",
1812
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1813
+ "dataType": "sc:Text",
1814
+ "source": {
1815
+ "fileObject": {
1816
+ "@id": "file-train-00013-of-00018.parquet"
1817
+ },
1818
+ "extract": {
1819
+ "column": "category"
1820
+ }
1821
+ }
1822
+ }
1823
+ ]
1824
+ },
1825
+ {
1826
+ "@type": "cr:RecordSet",
1827
+ "@id": "train-00014-of-00018",
1828
+ "name": "train-00014-of-00018",
1829
+ "description": "train split, physical shard train/train-00014-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1830
+ "field": [
1831
+ {
1832
+ "@type": "cr:Field",
1833
+ "@id": "train-00014-of-00018/source",
1834
+ "name": "source",
1835
+ "description": "Reference garment image.",
1836
+ "dataType": "sc:ImageObject",
1837
+ "source": {
1838
+ "fileObject": {
1839
+ "@id": "file-train-00014-of-00018.parquet"
1840
+ },
1841
+ "extract": {
1842
+ "column": "source"
1843
+ },
1844
+ "transform": {
1845
+ "jsonPath": "bytes"
1846
+ }
1847
+ }
1848
+ },
1849
+ {
1850
+ "@type": "cr:Field",
1851
+ "@id": "train-00014-of-00018/mask",
1852
+ "name": "mask",
1853
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1854
+ "dataType": "sc:ImageObject",
1855
+ "source": {
1856
+ "fileObject": {
1857
+ "@id": "file-train-00014-of-00018.parquet"
1858
+ },
1859
+ "extract": {
1860
+ "column": "mask"
1861
+ },
1862
+ "transform": {
1863
+ "jsonPath": "bytes"
1864
+ }
1865
+ }
1866
+ },
1867
+ {
1868
+ "@type": "cr:Field",
1869
+ "@id": "train-00014-of-00018/target",
1870
+ "name": "target",
1871
+ "description": "Person wearing the reference garment (target image).",
1872
+ "dataType": "sc:ImageObject",
1873
+ "source": {
1874
+ "fileObject": {
1875
+ "@id": "file-train-00014-of-00018.parquet"
1876
+ },
1877
+ "extract": {
1878
+ "column": "target"
1879
+ },
1880
+ "transform": {
1881
+ "jsonPath": "bytes"
1882
+ }
1883
+ }
1884
+ },
1885
+ {
1886
+ "@type": "cr:Field",
1887
+ "@id": "train-00014-of-00018/caption",
1888
+ "name": "caption",
1889
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1890
+ "dataType": "sc:Text",
1891
+ "source": {
1892
+ "fileObject": {
1893
+ "@id": "file-train-00014-of-00018.parquet"
1894
+ },
1895
+ "extract": {
1896
+ "column": "caption"
1897
+ }
1898
+ }
1899
+ },
1900
+ {
1901
+ "@type": "cr:Field",
1902
+ "@id": "train-00014-of-00018/category",
1903
+ "name": "category",
1904
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1905
+ "dataType": "sc:Text",
1906
+ "source": {
1907
+ "fileObject": {
1908
+ "@id": "file-train-00014-of-00018.parquet"
1909
+ },
1910
+ "extract": {
1911
+ "column": "category"
1912
+ }
1913
+ }
1914
+ }
1915
+ ]
1916
+ },
1917
+ {
1918
+ "@type": "cr:RecordSet",
1919
+ "@id": "train-00015-of-00018",
1920
+ "name": "train-00015-of-00018",
1921
+ "description": "train split, physical shard train/train-00015-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
1922
+ "field": [
1923
+ {
1924
+ "@type": "cr:Field",
1925
+ "@id": "train-00015-of-00018/source",
1926
+ "name": "source",
1927
+ "description": "Reference garment image.",
1928
+ "dataType": "sc:ImageObject",
1929
+ "source": {
1930
+ "fileObject": {
1931
+ "@id": "file-train-00015-of-00018.parquet"
1932
+ },
1933
+ "extract": {
1934
+ "column": "source"
1935
+ },
1936
+ "transform": {
1937
+ "jsonPath": "bytes"
1938
+ }
1939
+ }
1940
+ },
1941
+ {
1942
+ "@type": "cr:Field",
1943
+ "@id": "train-00015-of-00018/mask",
1944
+ "name": "mask",
1945
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
1946
+ "dataType": "sc:ImageObject",
1947
+ "source": {
1948
+ "fileObject": {
1949
+ "@id": "file-train-00015-of-00018.parquet"
1950
+ },
1951
+ "extract": {
1952
+ "column": "mask"
1953
+ },
1954
+ "transform": {
1955
+ "jsonPath": "bytes"
1956
+ }
1957
+ }
1958
+ },
1959
+ {
1960
+ "@type": "cr:Field",
1961
+ "@id": "train-00015-of-00018/target",
1962
+ "name": "target",
1963
+ "description": "Person wearing the reference garment (target image).",
1964
+ "dataType": "sc:ImageObject",
1965
+ "source": {
1966
+ "fileObject": {
1967
+ "@id": "file-train-00015-of-00018.parquet"
1968
+ },
1969
+ "extract": {
1970
+ "column": "target"
1971
+ },
1972
+ "transform": {
1973
+ "jsonPath": "bytes"
1974
+ }
1975
+ }
1976
+ },
1977
+ {
1978
+ "@type": "cr:Field",
1979
+ "@id": "train-00015-of-00018/caption",
1980
+ "name": "caption",
1981
+ "description": "Garment description generated with Gemini 2.0 Flash.",
1982
+ "dataType": "sc:Text",
1983
+ "source": {
1984
+ "fileObject": {
1985
+ "@id": "file-train-00015-of-00018.parquet"
1986
+ },
1987
+ "extract": {
1988
+ "column": "caption"
1989
+ }
1990
+ }
1991
+ },
1992
+ {
1993
+ "@type": "cr:Field",
1994
+ "@id": "train-00015-of-00018/category",
1995
+ "name": "category",
1996
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
1997
+ "dataType": "sc:Text",
1998
+ "source": {
1999
+ "fileObject": {
2000
+ "@id": "file-train-00015-of-00018.parquet"
2001
+ },
2002
+ "extract": {
2003
+ "column": "category"
2004
+ }
2005
+ }
2006
+ }
2007
+ ]
2008
+ },
2009
+ {
2010
+ "@type": "cr:RecordSet",
2011
+ "@id": "train-00016-of-00018",
2012
+ "name": "train-00016-of-00018",
2013
+ "description": "train split, physical shard train/train-00016-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
2014
+ "field": [
2015
+ {
2016
+ "@type": "cr:Field",
2017
+ "@id": "train-00016-of-00018/source",
2018
+ "name": "source",
2019
+ "description": "Reference garment image.",
2020
+ "dataType": "sc:ImageObject",
2021
+ "source": {
2022
+ "fileObject": {
2023
+ "@id": "file-train-00016-of-00018.parquet"
2024
+ },
2025
+ "extract": {
2026
+ "column": "source"
2027
+ },
2028
+ "transform": {
2029
+ "jsonPath": "bytes"
2030
+ }
2031
+ }
2032
+ },
2033
+ {
2034
+ "@type": "cr:Field",
2035
+ "@id": "train-00016-of-00018/mask",
2036
+ "name": "mask",
2037
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
2038
+ "dataType": "sc:ImageObject",
2039
+ "source": {
2040
+ "fileObject": {
2041
+ "@id": "file-train-00016-of-00018.parquet"
2042
+ },
2043
+ "extract": {
2044
+ "column": "mask"
2045
+ },
2046
+ "transform": {
2047
+ "jsonPath": "bytes"
2048
+ }
2049
+ }
2050
+ },
2051
+ {
2052
+ "@type": "cr:Field",
2053
+ "@id": "train-00016-of-00018/target",
2054
+ "name": "target",
2055
+ "description": "Person wearing the reference garment (target image).",
2056
+ "dataType": "sc:ImageObject",
2057
+ "source": {
2058
+ "fileObject": {
2059
+ "@id": "file-train-00016-of-00018.parquet"
2060
+ },
2061
+ "extract": {
2062
+ "column": "target"
2063
+ },
2064
+ "transform": {
2065
+ "jsonPath": "bytes"
2066
+ }
2067
+ }
2068
+ },
2069
+ {
2070
+ "@type": "cr:Field",
2071
+ "@id": "train-00016-of-00018/caption",
2072
+ "name": "caption",
2073
+ "description": "Garment description generated with Gemini 2.0 Flash.",
2074
+ "dataType": "sc:Text",
2075
+ "source": {
2076
+ "fileObject": {
2077
+ "@id": "file-train-00016-of-00018.parquet"
2078
+ },
2079
+ "extract": {
2080
+ "column": "caption"
2081
+ }
2082
+ }
2083
+ },
2084
+ {
2085
+ "@type": "cr:Field",
2086
+ "@id": "train-00016-of-00018/category",
2087
+ "name": "category",
2088
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
2089
+ "dataType": "sc:Text",
2090
+ "source": {
2091
+ "fileObject": {
2092
+ "@id": "file-train-00016-of-00018.parquet"
2093
+ },
2094
+ "extract": {
2095
+ "column": "category"
2096
+ }
2097
+ }
2098
+ }
2099
+ ]
2100
+ },
2101
+ {
2102
+ "@type": "cr:RecordSet",
2103
+ "@id": "train-00017-of-00018",
2104
+ "name": "train-00017-of-00018",
2105
+ "description": "train split, physical shard train/train-00017-of-00018.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
2106
+ "field": [
2107
+ {
2108
+ "@type": "cr:Field",
2109
+ "@id": "train-00017-of-00018/source",
2110
+ "name": "source",
2111
+ "description": "Reference garment image.",
2112
+ "dataType": "sc:ImageObject",
2113
+ "source": {
2114
+ "fileObject": {
2115
+ "@id": "file-train-00017-of-00018.parquet"
2116
+ },
2117
+ "extract": {
2118
+ "column": "source"
2119
+ },
2120
+ "transform": {
2121
+ "jsonPath": "bytes"
2122
+ }
2123
+ }
2124
+ },
2125
+ {
2126
+ "@type": "cr:Field",
2127
+ "@id": "train-00017-of-00018/mask",
2128
+ "name": "mask",
2129
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
2130
+ "dataType": "sc:ImageObject",
2131
+ "source": {
2132
+ "fileObject": {
2133
+ "@id": "file-train-00017-of-00018.parquet"
2134
+ },
2135
+ "extract": {
2136
+ "column": "mask"
2137
+ },
2138
+ "transform": {
2139
+ "jsonPath": "bytes"
2140
+ }
2141
+ }
2142
+ },
2143
+ {
2144
+ "@type": "cr:Field",
2145
+ "@id": "train-00017-of-00018/target",
2146
+ "name": "target",
2147
+ "description": "Person wearing the reference garment (target image).",
2148
+ "dataType": "sc:ImageObject",
2149
+ "source": {
2150
+ "fileObject": {
2151
+ "@id": "file-train-00017-of-00018.parquet"
2152
+ },
2153
+ "extract": {
2154
+ "column": "target"
2155
+ },
2156
+ "transform": {
2157
+ "jsonPath": "bytes"
2158
+ }
2159
+ }
2160
+ },
2161
+ {
2162
+ "@type": "cr:Field",
2163
+ "@id": "train-00017-of-00018/caption",
2164
+ "name": "caption",
2165
+ "description": "Garment description generated with Gemini 2.0 Flash.",
2166
+ "dataType": "sc:Text",
2167
+ "source": {
2168
+ "fileObject": {
2169
+ "@id": "file-train-00017-of-00018.parquet"
2170
+ },
2171
+ "extract": {
2172
+ "column": "caption"
2173
+ }
2174
+ }
2175
+ },
2176
+ {
2177
+ "@type": "cr:Field",
2178
+ "@id": "train-00017-of-00018/category",
2179
+ "name": "category",
2180
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
2181
+ "dataType": "sc:Text",
2182
+ "source": {
2183
+ "fileObject": {
2184
+ "@id": "file-train-00017-of-00018.parquet"
2185
+ },
2186
+ "extract": {
2187
+ "column": "category"
2188
+ }
2189
+ }
2190
+ }
2191
+ ]
2192
+ },
2193
+ {
2194
+ "@type": "cr:RecordSet",
2195
+ "@id": "validation-00000-of-00001",
2196
+ "name": "validation-00000-of-00001",
2197
+ "description": "validation split, physical shard validation/validation-00000-of-00001.parquet. Concatenate all record sets with the same split prefix to obtain the complete split.",
2198
+ "field": [
2199
+ {
2200
+ "@type": "cr:Field",
2201
+ "@id": "validation-00000-of-00001/source",
2202
+ "name": "source",
2203
+ "description": "Reference garment image.",
2204
+ "dataType": "sc:ImageObject",
2205
+ "source": {
2206
+ "fileObject": {
2207
+ "@id": "file-validation-00000-of-00001.parquet"
2208
+ },
2209
+ "extract": {
2210
+ "column": "source"
2211
+ },
2212
+ "transform": {
2213
+ "jsonPath": "bytes"
2214
+ }
2215
+ }
2216
+ },
2217
+ {
2218
+ "@type": "cr:Field",
2219
+ "@id": "validation-00000-of-00001/mask",
2220
+ "name": "mask",
2221
+ "description": "Person image with the target garment region blacked out, used as the masked-person conditioning input.",
2222
+ "dataType": "sc:ImageObject",
2223
+ "source": {
2224
+ "fileObject": {
2225
+ "@id": "file-validation-00000-of-00001.parquet"
2226
+ },
2227
+ "extract": {
2228
+ "column": "mask"
2229
+ },
2230
+ "transform": {
2231
+ "jsonPath": "bytes"
2232
+ }
2233
+ }
2234
+ },
2235
+ {
2236
+ "@type": "cr:Field",
2237
+ "@id": "validation-00000-of-00001/target",
2238
+ "name": "target",
2239
+ "description": "Person wearing the reference garment (target image).",
2240
+ "dataType": "sc:ImageObject",
2241
+ "source": {
2242
+ "fileObject": {
2243
+ "@id": "file-validation-00000-of-00001.parquet"
2244
+ },
2245
+ "extract": {
2246
+ "column": "target"
2247
+ },
2248
+ "transform": {
2249
+ "jsonPath": "bytes"
2250
+ }
2251
+ }
2252
+ },
2253
+ {
2254
+ "@type": "cr:Field",
2255
+ "@id": "validation-00000-of-00001/caption",
2256
+ "name": "caption",
2257
+ "description": "Garment description generated with Gemini 2.0 Flash.",
2258
+ "dataType": "sc:Text",
2259
+ "source": {
2260
+ "fileObject": {
2261
+ "@id": "file-validation-00000-of-00001.parquet"
2262
+ },
2263
+ "extract": {
2264
+ "column": "caption"
2265
+ }
2266
+ }
2267
+ },
2268
+ {
2269
+ "@type": "cr:Field",
2270
+ "@id": "validation-00000-of-00001/category",
2271
+ "name": "category",
2272
+ "description": "One of 14 coarse labels obtained by deterministic keyword matching over Gemini captions; not a visual cluster ID.",
2273
+ "dataType": "sc:Text",
2274
+ "source": {
2275
+ "fileObject": {
2276
+ "@id": "file-validation-00000-of-00001.parquet"
2277
+ },
2278
+ "extract": {
2279
+ "column": "category"
2280
+ }
2281
+ }
2282
+ }
2283
+ ]
2284
+ }
2285
+ ]
2286
+ }