TYUST-RAICOM2026-CAIP

WeatherNet-RAICOM2026

Official solution for the RAICOM 2026 CAIP Weather Classification Challenge

🏆 Competition Achievement

🥉 Provincial Third Prize — RAICOM Cup 2026 (睿抗杯)
CAIP AI Algorithm Optimization Track

The proposed WeatherNet-RAICOM2026 solution achieved Third Prize (Provincial Level) in the RAICOM 2026 Developer Competition – CAIP AI Algorithm Optimization Track.

---

Overview

WeatherNet-RAICOM2026 is the solution developed for the RAICOM 2026 Developer Competition – CAIP (AI Algorithm Optimization Track).

The task aims to classify weather conditions from images into four categories:

• ☀️ Sunny
• ☁️ Cloudy
• 🌧️ Rainy
• ❄️ Snowy

This repository provides the trained model weights and inference implementation based entirely on PyTorch and Torchvision.

---

Model Architecture

The final solution adopts a lightweight ensemble framework consisting of two complementary CNN backbones:

The predictions from both models are fused using weighted soft-voting.

---

Inference Pipeline

The inference pipeline consists of several optimization strategies:

• Dual-model ensemble
• Multi-scale inference
• Horizontal Flip Test-Time Augmentation (TTA)
• Weighted probability fusion
• Macro-F1 score bias calibration

The entire inference pipeline is implemented using only torchvision, without additional third-party model libraries.

---

Project Structure

.
├── main.py
├── results
│   ├── ev2s.pth
│   └── cvt.pth
└── README.md