Back to Projects

VALORANT Round Outcome Prediction

With NAIST, I used TimeSformer to predict VALORANT round outcomes from minimap video.

Open External Link
Prediction Accuracy
80.55%
Round Videos
21,228
Published At
IEEE CoG

Overview

I collaborated with NAIST on predicting VALORANT round outcomes from minimap video. TimeSformer learned from 21,228 round videos drawn from 1,376 tournament videos and reached 80.55% prediction accuracy.

The minimap represents player movement and team coordination, so I used it as the input to the computer-vision approach. The model reached 80.55% accuracy on round-outcome prediction.

What I Built

  • TimeSformer

    I used TimeSformer to model temporal and spatial patterns in minimap video sequences.

  • Tournament Dataset

    I built the training and validation dataset from 1,376 tournament videos, including 21,228 round videos.

  • Round-Outcome Prediction

    The model reached 80.55% accuracy on round-outcome prediction from minimap video.

  • Minimap Pipeline

    I processed minimap footage to study player movement and team coordination from video.

Results

  • Published at IEEE Conference on Games 2025.
  • Applied video transformers to esports analytics in collaboration with NAIST.
  • Produced a dataset for follow-on research.

Collaboration

Conducted with the Nara Institute of Science and Technology (NAIST).