---
datasets:
- Vigen1/hypersim-mini
tags:
- computer-vision
- segmentation
task_categories:
- image-segmentation
pretty_name: Hypersim Minimal (RGB + Semantic Mapped)
license: other
dataset_info:
  features:
  - name: image
    dtype: image
  - name: mask
    dtype: image
  - name: scene
    dtype: string
  - name: camera
    dtype: string
  - name: frame
    dtype: string
  splits:
  - name: train
    num_bytes: 384739491.02400005
    num_examples: 4788
  - name: validation
    num_bytes: 42617139.0
    num_examples: 532
  download_size: 423207114
  dataset_size: 427356630.02400005
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
---

# Hypersim Minimal (RGB + Semantic Mapped)

This dataset is a minimal extraction from **Hypersim**:
- **RGB**: `preview` (either preview JPGs or HDR color.hdf5 tonemapped to PNG)
- **Semantic labels**: mapped to uint8 PNG using `clip40to39`

## Columns
- `image` — RGB image
- `mask` — segmentation mask (uint8)
- `scene`, `camera`, `frame` — identifiers

## Splits
- `train`
- `validation` (ratio: 0.1)

> Note: You are responsible for complying with the original Hypersim license/terms.