# Dataset for training segmentation on mobile game screens: 6.9k Clash of Clans frames, 1.9k classes — retained record

> Retained reddit record from reddit:deeplearning.

Canonical: https://highsignal.app/data/records/9991acc2cdd53839

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:deeplearning)  published 2026-11-01 05:23 ingested 2026-11-02 02:38

[original source ↗](https://www.reddit.com/r/deeplearning/comments/1xkuyk3/dataset_for_training_segmentation_on_mobile_game/)

## retained text



```
Every object on these screens is labelled as a polygon of its visible pixels, from buildings and walls down to troops, heroes, the army bar and HUD text. Frames cover the scout screen, the battle and the end of battle screen.



6,909 frames, about 4.2M masks, YOLO seg and COCO. Val is split by base.



Free sample and viewer: [https://huggingface.co/spaces/nihatxp/coc-dataset-viewer](https://huggingface.co/spaces/nihatxp/coc-dataset-viewer)

Full set is paid, details on the Hugging Face page.
```



document fetched 2026-11-02 00:23 · canonical https://reddit.com/r/deeplearning/comments/1xkuyk3/dataset_for_training_segmentation_on_mobile_game
