windows can't convert
==================== stitch_network done ====================
==================== warp_network start ====================
[12/22/2025-00:29:18] [TRT] [I] [MemUsageChange] Init CUDA: CPU +81, GPU +0, now: CPU 18292, GPU 1166 (MiB)
[12/22/2025-00:29:22] [TRT] [I] [MemUsageChange] Init builder kernel library: CPU +1435, GPU +268, now: CPU 21018, GPU 1434 (MiB)
[12/22/2025-00:29:52] [TRT] [E] 6: [libLoader.cpp::nvinfer1::rt::DynamicLibrary::Impl::Impl::290] Error Code 6: Internal Error (Unable to load library: ./checkpoints/ditto_onnx\libgrid_sample_3d_plugin.so)
[12/22/2025-00:29:52] [TRT] [I] ----------------------------------------------------------------
[12/22/2025-00:29:52] [TRT] [I] Input filename: ./checkpoints/ditto_onnx/warp_network.onnx
[12/22/2025-00:29:52] [TRT] [I] ONNX IR version: 0.0.10
[12/22/2025-00:29:52] [TRT] [I] Opset version: 17
[12/22/2025-00:29:52] [TRT] [I] Producer name: pytorch
[12/22/2025-00:29:52] [TRT] [I] Producer version: 2.4.0
[12/22/2025-00:29:52] [TRT] [I] Domain:
[12/22/2025-00:29:52] [TRT] [I] Model version: 0
[12/22/2025-00:29:52] [TRT] [I] Doc string:
[12/22/2025-00:29:52] [TRT] [I] ----------------------------------------------------------------
[12/22/2025-00:29:52] [TRT] [W] onnx2trt_utils.cpp:374: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
[12/22/2025-00:29:52] [TRT] [I] No importer registered for op: GridSample3D. Attempting to import as plugin.
[12/22/2025-00:29:52] [TRT] [I] Searching for plugin: GridSample3D, plugin_version: 1, plugin_namespace:
[12/22/2025-00:29:52] [TRT] [E] 3: getPluginCreator could not find plugin: GridSample3D version: 1
[12/22/2025-00:29:52] [TRT] [E] ModelImporter.cpp:771: While parsing node number 11 [GridSample3D -> "/dense_motion_network/GridSample_output_0"]:
[12/22/2025-00:29:52] [TRT] [E] ModelImporter.cpp:772: --- Begin node ---
[12/22/2025-00:29:52] [TRT] [E] ModelImporter.cpp:773: input: "/dense_motion_network/Reshape_3_output_0"
input: "/dense_motion_network/Reshape_4_output_0"
output: "/dense_motion_network/GridSample_output_0"
name: "/dense_motion_network/GridSample"
op_type: "GridSample3D"
attribute {
name: "name"
s: "GridSample3D"
type: STRING
}
attribute {
name: "version"
i: 1
type: INT
}
attribute {
name: "namespace"
s: ""
type: STRING
}
[12/22/2025-00:29:52] [TRT] [E] ModelImporter.cpp:774: --- End node ---
[12/22/2025-00:29:52] [TRT] [E] ModelImporter.cpp:777: ERROR: builtin_op_importers.cpp:5404 In function importFallbackPluginImporter:
[8] Assertion failed: creator && "Plugin not found, are the plugin name, version, and namespace correct?"
Fail parsing ./checkpoints/ditto_onnx/warp_network.onnx
In node 11 (importFallbackPluginImporter): UNSUPPORTED_NODE: Assertion failed: creator && "Plugin not found, are the plugin name, version, and namespace correct?"
Traceback (most recent call last):
File "D:\ditto-talkinghead-main\scripts\cvt_onnx_to_trt.py", line 163, in
main(onnx_dir, trt_dir, grid_sample_plugin_file)
File "D:\ditto-talkinghead-main\scripts\cvt_onnx_to_trt.py", line 141, in main
onnx_to_trt_for_gridsample(onnx_file, trt_file, fp16, plugin_file=grid_sample_plugin_file)
File "D:\ditto-talkinghead-main\scripts\cvt_onnx_to_trt.py", line 53, in onnx_to_trt_for_gridsample
f"{error.code() = }\n{error.file() = }\n{error.func() = }\n{error.line() = }\n{error.local_function_stack_size() = }"
AttributeError: 'tensorrt.tensorrt.ParserError' object has no attribute 'local_function_stack_size'
这一连串的报错信息和系统弹窗终于彻底揭开了“真相”：你正在尝试在 Windows 系统上运行专为 Linux 编译的二进制文件。
从你提供的新报错日志和截图可以确认：
格式冲突：报错弹窗显示程序尝试加载 libgrid_sample_3d_plugin.so。在 Windows 上，.so 是无法运行的（Windows 对应的是 .dll），这直接导致了 0xc000012f（损坏的映像）系统错误。
算子缺失：日志明确指出 TensorRT 找不到 GridSample3D 算子的实现（getPluginCreator could not find plugin: GridSample3D）。这是因为负责实现这个算子的插件库（那个 .so 文件）加载失败了。
转换失败：由于 scripts/cvt_onnx_to_trt.py 在转换 warp_network.onnx 时无法找到该插件，导致模型解析（Parsing）断言失败。