Valorant Object Detection: Annotated Game Images for Teammates, Enemies, and Spikes

9,898 annotated images from the video game Valorant, split into training, validation, and test sets. The dataset, created by keremberke and hosted on Hugging Face, was last updated in January 2023. It contains bounding box labels for four object classes relevant to gameplay.

Use Cases

• Train object detection models to identify in-game characters based on the 'teammate' and 'enemy' labels.
• Develop models for tracking the game state by detecting 'planted spike' and 'dropped spike' objects.
• Benchmark model performance on a domain-specific dataset with a defined train/valid/test split.
• Create synthetic training data pipelines for game AI by leveraging the provided annotations.

Strengths

• Provides 9,898 total annotated images, offering a substantial volume for model training.
• Includes a predefined data split with 6,927 training, 1,983 validation, and 988 test images.
• Focuses on four specific, actionable object classes ('dropped spike', 'enemy', 'planted spike', 'teammate') for gameplay analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last updated 2023-01-27 13:45:00; freshness should be verified against current game versions.
• The source and collection methodology for the images are not described, which may affect understanding of potential biases.

Provenance

Source

keremberke via Hugging Face and Roboflow Universe.

Collection Method

Likely contains screenshots or extracted frames from the Valorant video game, manually or semi-automatically annotated.

Freshness

Last updated 2023-01-27 13:45:00