CandleLabAI/FLORA

---
license: cc-by-2.0
task_categories:
- text-to-image
- image-to-text
language:
- en
tags:
- art
size_categories:
- 1K<n<10K
---
# FLORA: Fashion Language Outfit Representation for Apparel Generation

This repository hosts the **FLORA** dataset as a Hugging Face dataset.

FLORA is introduced in the WACV 2026 paper:

> **Dressing the Imagination: A Dataset for AI-Powered Translation of Text into Fashion Outfits and A Novel NeRA Adapter for Enhanced Feature Adaptation**

<div align="center">

<a href="https://candlelabai.github.io/WACV2026-FLORA-Dataset-NeRA-Adapter/">
<img src="https://img.shields.io/badge/%F0%9F%8C%8E-Project%20Website-purple" />
</a>

<a href="https://arxiv.org/pdf/2411.13901">
<img src="https://img.shields.io/badge/%F0%9F%93%96%20ArXiv-Paper-b31b1b" />
</a>

<a href="WACV_PAPER_URL_HERE">
<img src="https://img.shields.io/badge/%F0%9F%8C%B8%20WACV%202026-Paper-553C9A" />
</a>

</div>

FLORA is a curated text–image dataset of fashion **sketches** paired with rich, expert-level textual descriptions, designed for **text-to-fashion outfit generation** and related multimodal research.

---

## 🌸 Overview

**FLORA (Fashion Language Outfit Representation for Apparel Generation)** contains:

- Fashion **sketch images**
- A structured, detailed description containing:
- Model presence & gender
- Pose description
- Outfit structure & materials
- Color details
- Accessories and styling cues
- Descriptions use fashion-industry terminology for:
- Garment type & silhouette
- Style and construction details
- Fabrics, textures, and patterns

The dataset is meant to:

- Enable **text-to-fashion–sketch generation**
- Support **vision–language research** focused on fashion
- Serve as a benchmark for **fine-tuning diffusion and other generative models** in the fashion domain

Example image–prompt pairs from the FLORA dataset:


---

## 📁 Repository Structure

The Hugging Face dataset contains:

- **Main training split**
- Loaded as `ds["training_dataset"]`
- 4,330 rows with images and corresponding rich textual descriptions

- **Extra CSV files**
- `test_set.csv` – prompts for **quantitative & qualitative testing**
- `additional_abstract_prompts.csv` – extra prompts for **stress-testing** and **generalization/evaluation**

Both CSVs provide prompts that are compatible with models trained on the main FLORA training set.

---

## 📥 How to Load the Dataset

### 1. Install Dependencies

```bash
pip install datasets huggingface_hub pandas
```

### 2. Load from Hugging Face

```bash
from datasets import load_dataset
from huggingface_hub import hf_hub_download
import pandas as pd

# Main image dataset
ds = load_dataset("CandleLabAI/FLORA")
train = ds["training_dataset"]

# Load the testing CSVs
test_set = hf_hub_download(
repo_id="CandleLabAI/FLORA",
repo_type="dataset",
filename="test_set.csv",
)
additional_abstract_prompts = hf_hub_download(
repo_id="CandleLabAI/FLORA",
repo_type="dataset",
filename="additional_abstract_prompts.csv",
)

test_set = pd.read_csv(test_set)
additional_abstract_prompts = pd.read_csv(additional_abstract_prompts)
```

```train``` is a 🤗 Dataset object; ```test_set``` and ```additional_abstract_prompts``` are loaded as pandas DataFrames.

---

## 📜 Citation

If you use **FLORA** or **NeRA adapter** in your research, please cite:
> Dressing the Imagination: A Dataset for AI-Powered Translation of Text into Fashion Outfits and A Novel NeRA Adapter for Enhanced Feature Adaptation
>
> WACV 2026

```bibtex
@inproceedings{Deshmukh_2026_WACV,
author = {Deshmukh, Gayatri and De, Somsubhra and Sehgal, Chirag and Gupta, Jishu Sen and Mittal, Sparsh},
title = {Dressing the Imagination: A Dataset for AI-Powered Translation of Text into Fashion Outfits and A Novel NeRA Adapter for Enhanced Feature Adaptation},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
month = {March},
year = {2026},
}
```
---

⚖️ License

The FLORA dataset is intended for research purposes.
Please see the ```LICENSE ``` file and/or the Hugging Face dataset card for full terms.