nirmalendu01/person-background-dataset

Person-Background Dataset

Diverse people placed in various scene backgrounds. Generated with FLUX.1-dev (people) and FLUX.1-Kontext (background editing).

How It Is Collected

The collect.py script:

1. Stage 1 – People: Generates 10 diverse people with FLUX.1-dev (different ethnicities, genders, ages).
2. Stage 2 – Backgrounds: Uses FLUX.1-Kontext img2img to edit each person into scene categories (forest, beach, office, etc.). Preserves person identity while changing only the background.

Columns

Column	Description
source_image	Original person image
edited_image	Person placed in the scene background
scene	Scene category (e.g. "forest", "beach")
scene_prompt	Edit prompt used
person_id	Person index (person_0 … person_9)

Dataset Stats

• Total rows: 3600
• People: 10 (360 samples each)
• Scenes: 18 (200 samples each)

Samples per scene

scene	count
forest	200
beach	200
mountain	200
desert	200
park	200
jungle	200
city street	200
office	200
classroom	200
library	200
living room	200
kitchen	200
church interior	200
mosque interior	200
temple interior	200
train station	200
market	200
shopping mall	200

Samples per person_id

person_id	count
person_0	360
person_1	360
person_2	360
person_3	360
person_4	360
person_5	360
person_6	360
person_7	360
person_8	360
person_9	360

Usage

from datasets import load_dataset

ds = load_dataset("nirmalendu01/person-background-dataset", split="train")
# Filter by scene
forest = ds.filter(lambda x: x["scene"] == "forest")
# Filter by person
person_0 = ds.filter(lambda x: x["person_id"] == "person_0")