Uncertainty-Guided Geometric Pseudo-label Self-Training for Monocular Non-cooperative Spacecraft Pose Estimation with ConvMamSPN / 基于不确定性引导的几何伪标签自训练方法用于跨域的单目非合作航天器姿态估计

• This project currently only includes the complete project structure and placeholder code, and the complete code will be released after the paper is accepted.
• 本项目目前仅包含完整的项目结构和占位代码，完整代码将在论文被接收后发布。
• This project currently only includes the complete project structure and placeholder code, and the complete code will be released after the paper is accepted.
• 本项目目前仅包含完整的项目结构和占位代码，完整代码将在论文被接收后发布。
• This project currently only includes the complete project structure and placeholder code, and the complete code will be released after the paper is accepted.
• 本项目目前仅包含完整的项目结构和占位代码，完整代码将在论文被接收后发布。

Environment Setup / 环境配置

Please set up the environment according to environment.yml:

conda env create -f environment.yml
conda activate mambapose

---

Dataset Preparation / 数据集准备

SPEED and SPEED+ are publicly available from official Stanford/SLAB pages. SPEED 和 SPEED+ 数据集可从官方 Stanford/SLAB 页面获取。

• Stanford SLAB project page (SPEED / SPEED+):
  https://slab.stanford.edu/projects/speed
• SPEED+ Stanford Digital Repository:
  https://purl.stanford.edu/wv398fc4383

Notes / 说明：

• This repository does not redistribute dataset files.
  本仓库不重新分发数据集文件。
• Please download from official sources and organize locally.
  请从官方地址下载并在本地组织数据。

---

Training ConvMamSPN / 训练 ConvMamSPN

Use Python entry files directly:

# source-domain training / 源域训练
python main_single_stage.py

# UGGP target-domain training / UGGP目标域训练
python train_uggp.py

---

UGGP Workflow (Actual) / UGGP真实流程

Current implementation / 当前实现：

1. Train source-domain model with main_single_stage.py.
   先使用 main_single_stage.py 训练源域模型。
2. Run UGGP training with train_uggp.py.
   再使用 train_uggp.py 进行 UGGP 训练。
3. Teacher model generates pseudo labels online in real time to supervise the student model.
   训练过程中由教师模型实时在线生成伪标签监督学生模型。

There is no separate offline pseudo-label generation stage. 当前流程没有离线伪标签预生成步骤。

---

Current directory structure

ConvMamSPN_UDA/
├─ AM.yaml
├─ EffcientMethod/
├─ loaders/
├─ log/
├─ losses/
├─ main_single_stage.py
├─ main_single_stage_swisscube.py
├─ models/
├─ results/
├─ src/
├─ test_single_stage.py
├─ test_swisscube_adi.py
├─ train.json
├─ train_uggp.py
├─ txt2csv.py
├─ utils/
├─ visual/
├─ 主干消融.log
├─ 主干消融_CLEAN.csv
├─ 主干消融_CLEAN.txt
├─ 批量测试.py
├─ 批量运行.py
└─ 批量运行stu.py