EffiRLD 高效無人機影像水稻倒伏辨識模型 - EffiRLD_lodg01_OpenPack.zip

EffiRLD_lodg01_OpenPack.zip

EffiRLD 模型推論套件（版本 lodg01）

完整且自足的推論套件，解壓縮後為 OpenPack 資料夾，只需安裝 PyTorch、NumPy 與 Pillow 即可執行。

內容

模型權重 checkpoint-best.pth（683,622 個參數，含 optimizer 狀態，約 8.5 MB）
推論套件 edanet_eca 與命令列工具：predict.py（單張、整個資料夾，或以滑動視窗推論大幅正射影像）、selftest.py（自我檢查）、evaluate_h5.py（在 HDF5 測試集上重現指標）
3 張測試集範例圖磚、標註與參考輸出（samples）
訓練設定與測試指標（configs）
環境建置檔：Dockerfile（Docker）、env/environment.yml（conda）、env/apptainer.def（Apptainer）、env/requirements-pinned.txt（Python venv，固定實測版本）
使用說明 README.md（英文）與 README.zh-TW.md（繁體中文）、LICENSE、NOTICE、CITATION.cff、MANIFEST.sha256

快速開始

也可依 README 第 2 節，以 Docker、conda 或 Apptainer 建立環境（含 NVIDIA GPU 版）。

效能（獨立測試集，2,176 張 512×512 圖磚）：像素準確率 93.42%、mIoU 78.56%、倒伏稻 F1-score 85.17%。

檔案完整性

授權：Apache License 2.0（可商用）。再散布時請保留 LICENSE 與 NOTICE。

EffiRLD model package for inference (version lodg01)

A complete, self-contained inference package. It unpacks into an OpenPack folder and runs with only PyTorch, NumPy and Pillow installed.

Contents

Model weights checkpoint-best.pth (683,622 parameters, including the optimizer state, about 8.5 MB)
The edanet_eca inference package and command-line tools: predict.py (a single image, a whole folder, or sliding-window inference on large orthophotos), selftest.py (self-check) and evaluate_h5.py (reproduces the test metrics on an HDF5 test set)
Three test-set sample tiles with annotations and reference outputs (samples)
Training configuration and test metrics (configs)
Environment files: Dockerfile (Docker), env/environment.yml (conda), env/apptainer.def (Apptainer) and env/requirements-pinned.txt (Python venv, pinned to the tested versions)
Usage guides README.md (English) and README.zh-TW.md (Traditional Chinese), LICENSE, NOTICE, CITATION.cff and MANIFEST.sha256

Quick start

Docker, conda and Apptainer environments (including NVIDIA GPU builds) are described in section 2 of the README.

Performance (independent test set, 2,176 tiles of 512×512): pixel accuracy 93.42%, mIoU 78.56%, lodging F1-score 85.17%.

Integrity

After downloading, compare the SHA-256 hash of the zip with SHA256SUMS.txt on this page
After unzipping, run sha256sum -c MANIFEST.sha256 inside the OpenPack folder to check every file in the package

License: Apache License 2.0 (commercial use permitted). Keep LICENSE and NOTICE when redistributing.