SuperFlickr-30K-LARGE-Remastered

## **SuperFlickr-30K-LARGE-Remastered** **SuperFlickr-30K-LARGE-Remastered** is a remastered version of the Flickr30K dataset, where every image has been paired with newly generated, long-form descriptive captions using an automated **Qwen3.5 Multimodal** captioning pipeline. Instead of short human-written image descriptions, this dataset provides significantly richer scene-level annotations that describe objects, people, activities, spatial relationships, environments, lighting, and contextual details. The resulting captions offer substantially denser visual grounding for training and evaluating modern vision-language models. The dataset is derived from the original **Flickr30K** image collection introduced in *From Image Descriptions to Visual Denotations: New Similarity Metrics for Semantic Inference over Event Descriptions*. Every image has been re-captioned through a fully automated pipeline to generate detailed synthetic descriptions while preserving the original visual content. The dataset is intended for image captioning, multimodal pretraining, vision-language instruction tuning, retrieval, and dense image understanding tasks. ## Dataset Statistics | Property | Value | |-----------|-------| | Number of Samples | 31,783 | | Task | Image Captioning | | Caption Type | Long-form Synthetic Captions | | Caption Generator | Qwen3.5 Multimodal | | Dataset Format | Optimized Parquet | ## Dataset Structure Each sample contains the following fields: | Column | Type | Description | |---------|------|-------------| | `image` | Image | Original Flickr30K image | | `response` | String | Rich long-form caption generated by the Qwen3.5 Multimodal pipeline | Example: ```python sample = dataset[0] print(sample.keys()) # dict_keys([ # "image", # "response" # ]) ``` ## Loading the Dataset ```python from datasets import load_dataset dataset = load_dataset( "prithivMLmods/SuperFlickr-30K-LARGE-Remastered", split="train" ) ``` ## Example Usage ```python from datasets import load_dataset import matplotlib.pyplot as plt dataset = load_dataset( "prithivMLmods/SuperFlickr-30K-LARGE-Remastered", split="train" ) sample = dataset[0] image = sample["image"] caption = sample["response"] print(caption) plt.figure(figsize=(8, 8)) plt.imshow(image) plt.axis("off") plt.show() ``` ## Original Dataset This dataset is based on the following resources: - Flickr30K: https://aclanthology.org/Q14-1006/ - From Image Descriptions to Visual Denotations: New Similarity Metrics for Semantic Inference over Event Descriptions: https://shannon.cs.illinois.edu/DenotationGraph/ ## Citation If you use this remastered dataset in your research or projects, please cite: ```bibtex @dataset{SuperFlickr30KLargeRemastered2026, title={SuperFlickr-30K-LARGE-Remastered}, author={Prithiv Sakthi}, year={2026}, publisher={Hugging Face}, url={https://huggingface.co/prithivMLmods} } ``` Please also cite the original Flickr30K paper: ```bibtex @article{young2014flickr30k, title={From Image Descriptions to Visual Denotations: New Similarity Metrics for Semantic Inference over Event Descriptions}, author={Young, Peter and Lai, Alice and Hodosh, Micah and Hockenmaier, Julia}, journal={Transactions of the Association for Computational Linguistics}, volume={2}, pages={67--78}, year={2014} } ``` ## License This dataset is released under the **Apache-2.0 License**.