Data collection
We collect images from 42 webcam viewpoints around Mount Fuji at 30-minute intervals and align them with timestamped meteorological observations and forecasts.
Accepted to WACV 2026
FujiView
FujiView is a multimodal computer vision framework for Mount Fuji visibility estimation, combining webcam imagery with aligned meteorological data. Our WACV 2026 paper studies human-labeled five-class visibility prediction using visual models, weather features, and late-fusion learning; the public web app and dataset are coming soon.
Paper link available now • Web app + dataset release in progress
Methods at a Glance
A compact WACV-style overview
FujiView pairs computer vision with structured weather signals to model Mount Fuji visibility as a multimodal classification task.
Human-labeled visibility
Each image is manually labeled into five visibility categories: Perfect, Clear, Cloudy, Obscured, and Bad, creating supervision tailored to real scenic visibility rather than generic weather tags.
Multimodal modeling
Our pipeline compares image-based signals from YOLOv8 and MAE-style representations, then combines visual outputs with weather features in a late-fusion classifier.
Evaluation
We study the contribution of each modality through ablations and hyperparameter sweeps, targeting strong nowcasting performance while laying groundwork for short-term forecasting.
Dataset Scale
Current scale and release target
FujiView is being built as a large-scale scenic visibility benchmark, with both current collection statistics and projected release scale.
42
live webcams in the collection pipeline
113,000+
images collected so far
~26,000
manually labeled image-weather samples
300k+
forecast-aligned images from 40+ webcams in the released benchmark
Projected to exceed 320,000 total images during collection
The current paper reports over 113,000 collected images and approximately 26,000 manual labels, while the poster frames the broader public benchmark target as 300k+ forecast-aligned images from 40+ webcams.
Coming Soon
What’s next for FujiView
We’re preparing a public release of the FujiView experience and resources for researchers, developers, and weather / vision enthusiasts.
-
A public human-labeled dataset for Mount Fuji visibility
-
A web app for exploring visibility predictions and conditions
-
Foundations for broader environmental forecasting and vision tasks