目录第一步3. 清理旧构建产物 可选第2步. VS2022 x64 终端完整执行第3步 覆盖文件2、覆盖替换测试是否成功推理修改prroi_pooling_gpu.c 换成prroi_pooling_gpu.cpp修改模型路径三个步骤一步都不少第一步pytracking-master\ltr\external\PreciseRoIPoolingsetup.pyfrom setuptools import setup from torch.utils.cpp_extension import CUDAExtension, BuildExtension setup( nameprroi_pooling, ext_modules[ CUDAExtension( prroi_pooling, sources[ pytorch/prroi_pool/src/prroi_pooling_gpu.cpp, # 改后缀cpp pytorch/prroi_pool/src/prroi_pooling_gpu_impl.cu, ], include_dirs[ pytorch/prroi_pool/src ], ) ], cmdclass{ build_ext: BuildExtension } )prroi_pooling_gpu.c改名为prroi_pooling_gpu.cpp3. 清理旧构建产物 可选删除PreciseRoIPooling 目录下build文件夹C:\Users\Administrator\AppData\Local\torch_extensions整个文件夹第2步. VS2022 x64 终端完整执行conda activate pysd set DISTUTILS_USE_SDK1 set MSSdk1 cd F:\project\tracker_pix\pytracking-master\ltr\external\PreciseRoIPooling python setup.py build_ext --inplace第3步 覆盖文件从这个仓库根目录的src/拿到真正的两个大文件src/prroi_pooling_gpu_impl.cusrc/prroi_pooling_gpu_impl.cuh2、覆盖替换复制这两个真实文件覆盖到你的项目路径F:\project\tracker_pix\pytracking-master\ltr\external\PreciseRoIPooling\pytorch\prroi_pool\src\测试是否成功python -c import os; os.add_dll_directory(rC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\bin); os.add_dll_directory(rD:\soft\python\dam4\Lib\site-packages\torch\lib); import prroi_pooling; print(✅ 导入成功) ✅ 导入成功推理修改prroi_pooling_gpu.c 换成prroi_pooling_gpu.cpp_prroi_pooling load_extension( _prroi_pooling, [pjoin(root_dir, prroi_pooling_gpu.cpp), pjoin(root_dir, prroi_pooling_gpu_impl.cu)], verboseTrue )修改模型路径E:\project\track\pytracking-master\ltr\admin\loading.pycheckpoint_path