| --- |
| license: mit |
| task_categories: |
| - visual-question-answering |
| - image-classification |
| - reinforcement-learning |
| tags: |
| - rlhf |
| - rl |
| - preference-modeling |
| - vision-language |
| - multimodal |
| - gym |
| - game |
| - gameplay |
| --- |
| |
| # Ghost Hunter RLHF Dataset |
|
|
| This dataset contains screenshots captured during gameplay of "Ghost Hunter" (8-bit FPS). Each image corresponds to a moment when the player successfully destroyed a ghost with the precision auto-fire. The dataset is intended for reinforcement learning from human feedback (RLHF) tasks, such as training a preference model to distinguish between "good" and "bad" shots. |
|
|
|
|
| **The gym is located in the /gym/ folder so users can play Ghost Hunter by webXOS and make their own datasets.** |
|
|
| ## Dataset Structure |
|
|
| - `images/`: PNG frames captured at the moment of a kill. |
| - `metadata.json`: JSON array with per-frame metadata including round number, ghosts remaining, player position, kills in that shot, and timestamp. |
|
|
| ## Stats |
|
|
| - Total captures: 81 |
| - Rounds covered: 10 |
| - Time span: 42.2 seconds |
|
|
| ## Usage |
|
|
| Use this dataset to fine-tune vision-language models or as a reward model input for RLHF. Each image can be paired with the question: "Is this a good aim?" or similar. |
|
|
| ## License |
|
|
| MIT |
