---
license: mit
dataset_info:
  features:
  - name: frame_image
    dtype: image
  - name: frame_metadata
    dtype: string
---

# RandomDoomSamples-110M Dataset

This dataset contains random Doom gameplay frames generated using a random agent policy.

## Dataset Summary
- **Total Episodes**: 4,128+
- **Estimated Frames**: 15M+
- **Format**: PNG images + JSON metadata per frame
- **Size**: ~253GB

## Structure
```
episodes/
├── 0/
│   ├── frame_0000.png
│   ├── frame_0000.json
│   └── ...
├── 1/
│   └── ...
└── ...
```

## Upload Status
Dataset is being uploaded in chunks to optimize memory usage.

## Usage
```python
from datasets import load_dataset

# Load in streaming mode to avoid memory issues
dataset = load_dataset("invocation02/RandomDoomSamples-110M", streaming=True)
for example in dataset:
    # Process frames
    pass
```

## Data Collection
- **Game**: Doom (VizDoom environment)
- **Agent**: Random action policy
- **Resolution**: Game screenshots
- **Metadata**: Game state, actions, rewards