bev-project/scripts/check_env_detailed.sh

#!/bin/bash
# BEVFusion 详细环境检查脚本（参考README.md）

echo "=========================================="
echo "BEVFusion 环境依赖检查"
echo "参考：README.md Prerequisites"
echo "=========================================="
echo ""

PASS=0
FAIL=0

# 检查函数
check_pass() {
    echo "  ✓ $1"
    ((PASS++))
}

check_fail() {
    echo "  ✗ $1"
    ((FAIL++))
}

check_warn() {
    echo "  ⚠ $1"
}

# 1. Python版本检查
echo "[1/12] Python 版本 (需要 >= 3.8, < 3.9)"
PYTHON_VERSION=$(python3 --version 2>&1 | awk '{print $2}')
PYTHON_MAJOR=$(echo $PYTHON_VERSION | cut -d. -f1)
PYTHON_MINOR=$(echo $PYTHON_VERSION | cut -d. -f2)

if [ "$PYTHON_MAJOR" -eq 3 ] && [ "$PYTHON_MINOR" -eq 8 ]; then
    check_pass "Python $PYTHON_VERSION"
else
    check_warn "Python $PYTHON_VERSION (推荐3.8.x)"
fi

# 2. PyTorch版本检查
echo ""
echo "[2/12] PyTorch 版本 (需要 >= 1.9, <= 1.10.2)"
PYTORCH_VERSION=$(python3 -c "import torch; print(torch.__version__)" 2>/dev/null)
if [ $? -eq 0 ]; then
    check_pass "PyTorch $PYTORCH_VERSION"
    
    # 检查CUDA
    CUDA_AVAILABLE=$(python3 -c "import torch; print(torch.cuda.is_available())" 2>/dev/null)
    if [ "$CUDA_AVAILABLE" = "True" ]; then
        GPU_COUNT=$(python3 -c "import torch; print(torch.cuda.device_count())" 2>/dev/null)
        check_pass "CUDA 可用，GPU数量: $GPU_COUNT"
    else
        check_fail "CUDA 不可用"
    fi
else
    check_fail "PyTorch 未安装"
fi

# 3. torchpack检查
echo ""
echo "[3/12] torchpack (必需)"
TORCHPACK_VERSION=$(python3 -c "import torchpack; print(torchpack.__version__)" 2>/dev/null)
if [ $? -eq 0 ]; then
    check_pass "torchpack $TORCHPACK_VERSION"
else
    check_fail "torchpack 未安装"
    echo "       安装: pip install torchpack"
fi

# 4. mmcv检查
echo ""
echo "[4/12] mmcv (需要 = 1.4.0)"
MMCV_VERSION=$(python3 -c "import mmcv; print(mmcv.__version__)" 2>/dev/null)
if [ $? -eq 0 ]; then
    if [ "$MMCV_VERSION" = "1.4.0" ]; then
        check_pass "mmcv $MMCV_VERSION"
    else
        check_warn "mmcv $MMCV_VERSION (推荐1.4.0)"
    fi
else
    check_fail "mmcv 未安装"
    echo "       安装: pip install mmcv==1.4.0 mmcv-full==1.4.0"
fi

# 5. mmdet检查
echo ""
echo "[5/12] mmdetection (需要 = 2.20.0)"
MMDET_VERSION=$(python3 -c "import mmdet; print(mmdet.__version__)" 2>/dev/null)
if [ $? -eq 0 ]; then
    if [ "$MMDET_VERSION" = "2.20.0" ]; then
        check_pass "mmdetection $MMDET_VERSION"
    else
        check_warn "mmdetection $MMDET_VERSION (推荐2.20.0)"
    fi
else
    check_fail "mmdetection 未安装"
    echo "       安装: pip install mmdet==2.20.0"
fi

# 6. OpenMPI和mpi4py检查
echo ""
echo "[6/12] OpenMPI 和 mpi4py (torchpack需要)"
if command -v mpirun &> /dev/null; then
    OMPI_VERSION=$(mpirun --version 2>&1 | head -n 1)
    check_pass "OpenMPI: $OMPI_VERSION"
else
    check_warn "OpenMPI 未安装（可选）"
fi

MPI4PY_VERSION=$(python3 -c "import mpi4py; print(mpi4py.__version__)" 2>/dev/null)
if [ $? -eq 0 ]; then
    check_pass "mpi4py $MPI4PY_VERSION"
else
    check_warn "mpi4py 未安装（可选）"
fi

# 7. Pillow检查
echo ""
echo "[7/12] Pillow (需要 = 8.4.0)"
PILLOW_VERSION=$(python3 -c "from PIL import Image; import PIL; print(PIL.__version__)" 2>/dev/null)
if [ $? -eq 0 ]; then
    if [ "$PILLOW_VERSION" = "8.4.0" ]; then
        check_pass "Pillow $PILLOW_VERSION"
    else
        check_warn "Pillow $PILLOW_VERSION (推荐8.4.0)"
    fi
else
    check_fail "Pillow 未安装"
fi

# 8. nuscenes-devkit检查
echo ""
echo "[8/12] nuscenes-devkit (必需)"
python3 -c "from nuscenes import NuScenes" 2>/dev/null
if [ $? -eq 0 ]; then
    check_pass "nuscenes-devkit 已安装"
else
    check_fail "nuscenes-devkit 未安装"
    echo "       安装: pip install nuscenes-devkit"
fi

# 9. tqdm检查
echo ""
echo "[9/12] tqdm (必需)"
python3 -c "import tqdm" 2>/dev/null
if [ $? -eq 0 ]; then
    check_pass "tqdm 已安装"
else
    check_fail "tqdm 未安装"
    echo "       安装: pip install tqdm"
fi

# 10. 自定义CUDA算子检查
echo ""
echo "[10/12] 自定义CUDA算子 (需要运行 python setup.py develop)"
python3 -c "from mmdet3d.ops import bev_pool_v2" 2>/dev/null
if [ $? -eq 0 ]; then
    check_pass "BEV Pool 算子已编译"
else
    check_fail "BEV Pool 算子未编译"
    echo "       运行: cd /workspace/bevfusion && python3 setup.py develop"
fi

python3 -c "from mmdet3d.ops import Voxelization" 2>/dev/null
if [ $? -eq 0 ]; then
    check_pass "Voxelization 算子已编译"
else
    check_fail "Voxelization 算子未编译"
fi

python3 -c "from mmdet3d.ops.spconv import SparseConv3d" 2>/dev/null
if [ $? -eq 0 ]; then
    check_pass "Sparse Conv 算子已编译"
else
    check_fail "Sparse Conv 算子未编译"
fi

# 11. 数据集检查
echo ""
echo "[11/12] nuScenes 数据集"
DATA_ROOT="data/nuscenes"
if [ -d "$DATA_ROOT" ]; then
    check_pass "数据集目录存在"
    
    # 检查关键文件
    if [ -f "$DATA_ROOT/nuscenes_infos_train.pkl" ]; then
        check_pass "训练数据info文件"
    else
        check_fail "nuscenes_infos_train.pkl 缺失"
    fi
    
    if [ -f "$DATA_ROOT/nuscenes_infos_val.pkl" ]; then
        check_pass "验证数据info文件"
    else
        check_fail "nuscenes_infos_val.pkl 缺失"
    fi
    
    if [ -d "$DATA_ROOT/samples" ]; then
        SAMPLE_COUNT=$(find "$DATA_ROOT/samples" -type f | wc -l)
        check_pass "samples目录 ($SAMPLE_COUNT 文件)"
    else
        check_fail "samples目录缺失"
    fi
else
    check_fail "数据集目录不存在: $DATA_ROOT"
    echo "       参考README.md的Data Preparation部分"
fi

# 12. 预训练模型检查
echo ""
echo "[12/12] 预训练模型"
PRETRAIN_DIR="pretrained"
if [ -d "$PRETRAIN_DIR" ]; then
    check_pass "预训练模型目录存在"
    
    # 检查关键模型
    if [ -f "$PRETRAIN_DIR/swint-nuimages-pretrained.pth" ]; then
        SIZE=$(du -h "$PRETRAIN_DIR/swint-nuimages-pretrained.pth" | cut -f1)
        check_pass "swint-nuimages-pretrained.pth ($SIZE)"
    else
        check_warn "swint-nuimages-pretrained.pth 缺失"
    fi
    
    if [ -f "$PRETRAIN_DIR/lidar-only-det.pth" ]; then
        SIZE=$(du -h "$PRETRAIN_DIR/lidar-only-det.pth" | cut -f1)
        check_pass "lidar-only-det.pth ($SIZE)"
    else
        check_warn "lidar-only-det.pth 缺失"
    fi
else
    check_warn "预训练模型目录不存在"
    echo "       运行: ./tools/download_pretrained.sh"
fi

# 总结
echo ""
echo "=========================================="
echo "检查完成"
echo "通过: $PASS 项"
echo "失败: $FAIL 项"
echo "=========================================="

if [ $FAIL -eq 0 ]; then
    echo ""
    echo "✓ 环境配置完整，可以开始训练！"
    echo ""
    echo "训练命令示例："
    echo "  # 3D检测"
    echo "  torchpack dist-run -np 8 python3 tools/train.py \\"
    echo "    configs/nuscenes/det/transfusion/secfpn/camera+lidar/swint_v0p075/convfuser.yaml \\"
    echo "    --model.encoders.camera.backbone.init_cfg.checkpoint pretrained/swint-nuimages-pretrained.pth \\"
    echo "    --load_from pretrained/lidar-only-det.pth"
    echo ""
    echo "  # BEV分割"
    echo "  torchpack dist-run -np 8 python3 tools/train.py \\"
    echo "    configs/nuscenes/seg/fusion-bev256d2-lss.yaml \\"
    echo "    --model.encoders.camera.backbone.init_cfg.checkpoint pretrained/swint-nuimages-pretrained.pth"
    exit 0
else
    echo ""
    echo "⚠ 发现 $FAIL 个问题，请先解决后再训练"
    echo ""
    echo "常见解决方案："
    echo "1. 安装缺失的包："
    echo "   pip install torchpack mmcv==1.4.0 mmcv-full==1.4.0 mmdet==2.20.0"
    echo "   pip install nuscenes-devkit tqdm Pillow==8.4.0"
    echo ""
    echo "2. 编译自定义算子："
    echo "   cd /workspace/bevfusion"
    echo "   python3 setup.py develop"
    echo ""
    echo "3. 准备数据集："
    echo "   参考README.md的Data Preparation部分"
    exit 1
fi