Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2505.17328 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910964157251584 |
|---|---|
| author | Kline, Dylan |
| author_facet | Kline, Dylan |
| contents | Foundational game-image encoders often overfit to game-specific visual styles, undermining performance on downstream tasks when applied to new games. We present a method that combines contrastive learning and domain-adversarial training to learn game-invariant visual features. By simultaneously encouraging similar content to cluster and discouraging game-specific cues via an adversarial domain classifier, our approach produces embeddings that generalize across diverse games. Experiments on the Bingsu game-image dataset (10,000 screenshots from 10 games) demonstrate that after only a few training epochs, our model's features no longer cluster by game, indicating successful invariance and potential for improved cross-game transfer (e.g., glitch detection) with minimal fine-tuning. This capability paves the way for more generalizable game vision models that require little to no retraining on new games. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_17328 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Game-invariant Features Through Contrastive and Domain-adversarial Learning Kline, Dylan Computer Vision and Pattern Recognition Foundational game-image encoders often overfit to game-specific visual styles, undermining performance on downstream tasks when applied to new games. We present a method that combines contrastive learning and domain-adversarial training to learn game-invariant visual features. By simultaneously encouraging similar content to cluster and discouraging game-specific cues via an adversarial domain classifier, our approach produces embeddings that generalize across diverse games. Experiments on the Bingsu game-image dataset (10,000 screenshots from 10 games) demonstrate that after only a few training epochs, our model's features no longer cluster by game, indicating successful invariance and potential for improved cross-game transfer (e.g., glitch detection) with minimal fine-tuning. This capability paves the way for more generalizable game vision models that require little to no retraining on new games. |
| title | Game-invariant Features Through Contrastive and Domain-adversarial Learning |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2505.17328 |