Major changes: - New frontend (platform/web/): Vite + React 18 + TypeScript + Tailwind - 4-module navigation: 数据送标 / 模型管理 / 车队管理 / 系统管理 - Data catalog with charts (DMS/ADAS/Lane 3-tab view) - Quality review workflow (标注质检): Good/Fine/Bad scoring with auto-advance - Audit enhancements: batch operations, rejection categories, Feishu notifications - Operation audit log (操作日志) - World model simulation studio (仿真工坊) - Dataset version management with snapshots and diff - ADAS 7-class dataset integration (138K images organized + compressed) - User management with Feishu integration and pagination - CRUD/search/filter on all pages, card layout redesign - PIL-optimized image overlay rendering - Auto-snapshot on build, in_review workflow stage - Removed embedded algorithm code (now in workspace)
1.6 KiB
1.6 KiB
lane_light + CPU PyTorch(UFLD 训练)
1. 激活环境
source /home/chengfanglu/miniconda3/etc/profile.d/conda.sh
conda activate lane_light
2. 已安装依赖(lane_light)
torch/torchvision(CPU 轮子,来自https://download.pytorch.org/whl/cpu)opencv-python,tqdm,tensorboard,addict,scikit-learn,pathspec(与requirements.txt对齐;sklearn包名在 pip 中为scikit-learn)
自检:
python -c "import torch; print('torch', torch.__version__, 'cuda=', torch.cuda.is_available())"
3. 数据与配置
- 默认数据根仍指向
lane0_reorganized/lane_training_pack(见configs/mufld_lane_culane.py)。 - CPU 建议使用
configs/mufld_lane_culane_cpu.py(batch_size=4,学习率与 warmup 已按 batch 相对 16 做了粗略缩放)。内存不够可改配置或命令行覆盖:
cd /home/chengfanglu/DATA/BK2/UFLD
python train.py configs/mufld_lane_culane_cpu.py --batch_size 2
4. 运行训练
cd /home/chengfanglu/DATA/BK2/UFLD
python train.py configs/mufld_lane_culane_cpu.py
说明:
train.py已改为在 无 CUDA 时使用cpu;原仓库中写死的CUDA_VISIBLE_DEVICES=1,2与.cuda()已去掉,避免 CPU 机直接报错。- 首次
pretrained=True会下载 ResNet 骨干权重,需联网。 - CPU 训练很慢,建议先用小
epoch/ 小batch_size做通路测试。
5. 可选:DataLoader num_workers
当前 data/dataloader.py 里 num_workers=8。若 CPU 内存紧张或不想多进程读盘,可自行把该值改小(例如 0 或 2)。