thb1314 / tensorrt-onnx-fasterrcnn-fpn-roialign Goto Github PK

我编译完成后，运行./pro fasterrcnn
报错：
[2022-02-24 19:15:07][info][app_fasterrcnn.cpp:122]:===================== test FasterRCNN FP32 rpn_backbone_resnet50 ==================================
[2022-02-24 19:15:07][info][trt_builder.cpp:471]:Compile FP32 Onnx Model 'rpn_backbone_resnet50.onnx'.
[2022-02-24 19:15:08][warn][trt_builder.cpp:33]:NVInfer: IElementWiseLayer with inputs 1529 and (Unnamed Layer* 500) [Shuffle]_output: first input has type Float but second input has type Int32.
[2022-02-24 19:15:08][error][trt_builder.cpp:30]:NVInfer: 4: [layers.cpp::validate::2292] Error Code 4: Internal Error (Add_370: operation SUM has incompatible input types Float and Int32)
[2022-02-24 19:15:08][error][trt_builder.cpp:30]:NVInfer: 4: [layers.cpp::validate::2292] Error Code 4: Internal Error (Add_370: operation SUM has incompatible input types Float and Int32)
[2022-02-24 19:15:08][error][trt_builder.cpp:30]:NVInfer: 4: [layers.cpp::validate::2292] Error Code 4: Internal Error (Add_370: operation SUM has incompatible input types Float and Int32)
[2022-02-24 19:15:08][error][trt_builder.cpp:30]:NVInfer: 4: [layers.cpp::validate::2292] Error Code 4: Internal Error (Add_370: operation SUM has incompatible input types Float and Int32)
[2022-02-24 19:15:08][error][trt_builder.cpp:30]:NVInfer: 4: [layers.cpp::validate::2292] Error Code 4: Internal Error (Add_370: operation SUM has incompatible input types Float and Int32)
[2022-02-24 19:15:08][error][trt_builder.cpp:30]:NVInfer: 4: [layers.cpp::validate::2292] Error Code 4: Internal Error (Add_370: operation SUM has incompatible input types Float and Int32)
[2022-02-24 19:15:08][error][trt_builder.cpp:30]:NVInfer: 4: [layers.cpp::validate::2292] Error Code 4: Internal Error (Add_370: operation SUM has incompatible input types Float and Int32)
[2022-02-24 19:15:08][error][trt_builder.cpp:30]:NVInfer: 4: [layers.cpp::validate::2292] Error Code 4: Internal Error (Add_370: operation SUM has incompatible input types Float and Int32)
pro: /home/xlj/005_ClionProjects/tensorrt-onnx-fasterrcnn-fpn-roialign-master/tensorrt_code/src/tensorRT/onnx_parser/ShapeTensor.hpp:57: int32_t onnx2trt::ShapeTensor::size() const: Assertion `sizeKnown()' failed.

我在运行x01export_FasterRCNN_onnx.py时，一直卡在onnxsim这一步，卡几个小时都没有结束运行，所以我就注释掉没有做simplifier，会不会是这一步导致的以上报错？可以可以发一份最终的new_header.onnx和rpn_backbone_resnet50.onnx到我的邮箱[email protected]，让我验证一下，万分感谢！

手写ai老哥，这个fasterrcnn tensorrt项目文件有点复杂，请问有独立的工程吗

大佬您好，我把这个sdk调试通过，能检测，batch_size为2时，返回检测结果都是第一张图片的，请问这个怎么解决呀？

onnx转化trt失败，我在群里没找到你，方便加我微信吗 15515650363

是否可以使用自己在pytorch faster-rcnn上训练得出的权重模型转onnx呢？具体需要怎么操作呢？谢谢！

identifier "bin_size_h" is undefined TensorRT E:\PycharmProjects\tensorrt-onnx-fasterrcnn-fpn-roialign-master\tensorrt_code\src\application\app_fasterrcnn\roialign.cu 188
identifier "fpn_lvl" is undefined TensorRT E:\PycharmProjects\tensorrt-onnx-fasterrcnn-fpn-roialign-master\tensorrt_code\src\application\app_fasterrcnn\fasterrcnn_decode.cu 187
identifier "output_val" is undefined TensorRT E:\PycharmProjects\tensorrt-onnx-fasterrcnn-fpn-roialign-master\tensorrt_code\src\application\app_fasterrcnn\roialign.cu 203

运行x07.py的时候遇到nodes in a graph must be topologically sorted,howerver input 'feature_0' of node :
name: RoiAlign_51 OpType: RoiAlign is not output of any previous nodes.

老哥如何解决

已start，还是关于fasterrcnn问题，方便留个联系方式吗

parsing onnx error rpn_backbone_resnet50:
(nbSpatialDims == kernelWeights.shape.nbDims - 2) && "The number of spatial dimensions and the kernel shape doesn't match up for the Conv operator."

您好，我在使用您的代码时出现了问题:
我将第4层改为了空洞卷积，其feature map由2525变为了5050，结果发现在检测大尺寸目标的时候，没有检测框，于是我将sptical_scales的1/32，调整为1/16，此时能够检测出大尺寸目标，但是预测框被缩小了一部分。同时，我在第4层特征图改为空洞卷积的基础上，将ASPP结构接在了第4层，此时也没有大尺寸目标的预测框，并且我将sptical_scales的1/32，调整为1/16也同样没有预测框。
我想问一下，faster rcnn.cpp文件中的RoiAlignImpl_Float()方法是跟python阶段的roialign方法是一样的吗，有没有差距呢

你好，在onnx转trt之后进行检测，得到的结果与onnx的预测结果不一致(置信度较低，有一些框未预测出)，请问除了阈值、参数这些，还可能是那些地方需要修改呢

请问可以使用这种方法部署torchvision下的maskrcnn么？需要修改哪些内容，如果我只需要目标不求mask的话，可以直接用么

大佬你好，我看你说的原则1和2：
原则1：reshape操作第一个值为-1，即设置为batch size大小不固定

原则2：针对一些tensorrt不友好的算子比如：原地赋值、nonzero算子，equal算子等尽可能以另外方式等价替代或者舍弃

分别对应您工程中的代码在哪一步（或者说是在哪里做了以上两个原则）呢？

你好，
我才用的是resnet101权重，但将onnx转为trt的时候出现了报错，能帮我解决一下吗：
[2022-12-09 17:06:23][info][app_fasterrcnn.cpp:110]:===================== test FasterRCNN FP32 rpn_backbone_resnet50_nopool ==================================
[2022-12-09 17:06:23][info][trt_builder.cpp:471]:Compile FP32 Onnx Model 'rpn_backbone_resnet50_nopool.onnx'.
[2022-12-09 17:06:32][error][trt_builder.cpp:30]:NVInfer: 9: [graphShapeAnalyzer.cpp::nvinfer1::builder::anonymous-namespace'::ShapeNodeRemover::throwIfError::1306] Error Code 9: Internal Error (Reshape_878: reshape changes volume ) [2022-12-09 17:06:32][error][trt_builder.cpp:30]:NVInfer: 9: [graphShapeAnalyzer.cpp::nvinfer1::builder::anonymous-namespace'::ShapeNodeRemover::throwIfError::1306] Error Code 9: Internal Error (Reshape_878: reshape changes volume
)
[2022-12-09 17:06:32][error][trt_builder.cpp:30]:NVInfer: 9: [graphShapeAnalyzer.cpp::nvinfer1::builder::anonymous-namespace'::ShapeNodeRemover::throwIfError::1306] Error Code 9: Internal Error (Reshape_878: reshape changes volume ) [2022-12-09 17:06:32][error][trt_builder.cpp:30]:NVInfer: 9: [graphShapeAnalyzer.cpp::nvinfer1::builder::anonymous-namespace'::ShapeNodeRemover::throwIfError::1306] Error Code 9: Internal Error (Reshape_878: reshape changes volume
)
[2022-12-09 17:06:32][error][trt_builder.cpp:30]:NVInfer: 9: [graphShapeAnalyzer.cpp::nvinfer1::builder::anonymous-namespace'::ShapeNodeRemover::throwIfError::1306] Error Code 9: Internal Error (Reshape_878: reshape changes volume ) [2022-12-09 17:06:32][error][trt_builder.cpp:30]:NVInfer: 9: [graphShapeAnalyzer.cpp::nvinfer1::builder::anonymous-namespace'::ShapeNodeRemover::throwIfError::1306] Error Code 9: Internal Error (Reshape_878: reshape changes volume
)
[2022-12-09 17:06:33][error][trt_builder.cpp:30]:NVInfer: 9: [graphShapeAnalyzer.cpp::nvinfer1::builder::anonymous-namespace'::ShapeNodeRemover::throwIfError::1306] Error Code 9: Internal Error (Reshape_878: reshape changes volume ) [2022-12-09 17:06:33][error][trt_builder.cpp:30]:NVInfer: 9: [graphShapeAnalyzer.cpp::nvinfer1::builder::anonymous-namespace'::ShapeNodeRemover::throwIfError::1306] Error Code 9: Internal Error (Reshape_878: reshape changes volume
)
[2022-12-09 17:06:33][error][trt_builder.cpp:30]:NVInfer: 9: [graphShapeAnalyzer.cpp::nvinfer1::builder::`anonymous-namespace'::ShapeNodeRemover::throwIfError::1306] Error Code 9: Internal Error (Reshape_878: reshape changes volume
)
[2022-12-09 17:06:33][error][trt_builder.cpp:30]:NVInfer: g:\chengc\project\tensorrt-onnx-fasterrcnn-fpn-roialign-master\tensorrt_code\src\tensorrt\onnx_parser\modelimporter.cpp:736: While parsing node number 363 [Add -> "onnx::Gather_1824"]:
[2022-12-09 17:06:33][error][trt_builder.cpp:30]:NVInfer: g:\chengc\project\tensorrt-onnx-fasterrcnn-fpn-roialign-master\tensorrt_code\src\tensorrt\onnx_parser\modelimporter.cpp:737: --- Begin node ---
[2022-12-09 17:06:33][error][trt_builder.cpp:30]:NVInfer: g:\chengc\project\tensorrt-onnx-fasterrcnn-fpn-roialign-master\tensorrt_code\src\tensorrt\onnx_parser\modelimporter.cpp:738: input: "onnx::Add_1800"
input: "onnx::Mul_1805"
output: "onnx::Gather_1824"
name: "Add_884"
op_type: "Add"

[2022-12-09 17:06:33][error][trt_builder.cpp:30]:NVInfer: g:\chengc\project\tensorrt-onnx-fasterrcnn-fpn-roialign-master\tensorrt_code\src\tensorrt\onnx_parser\modelimporter.cpp:739: --- End node ---
[2022-12-09 17:06:33][error][trt_builder.cpp:30]:NVInfer: g:\chengc\project\tensorrt-onnx-fasterrcnn-fpn-roialign-master\tensorrt_code\src\tensorrt\onnx_parser\modelimporter.cpp:742: ERROR: g:\chengc\project\tensorrt-onnx-fasterrcnn-fpn-roialign-master\tensorrt_code\src\tensorrt\onnx_parser\onnx2trt_utils.cpp:959 In function elementwiseHelper:
[8] Assertion failed: tensor_ptr->getDimensions().nbDims == maxNbDims && "Failed to broadcast tensors elementwise!"

这个应该是maskrcnn的部署吧

hello， thanks for your share. when run command ./pro fasterrcnn, i encounted one problem in function void InferImpl::forward(bool sync) with this line
bool execute_result = context->context_->enqueueV2(bindingsptr, context->stream_, nullptr);
, the details of this error is below,

[2022-06-02 17:01:11][error][trt_builder.cpp:30]:NVInfer: 1: [runner.cpp::execute::416] Error Code 1: Cuda Runtime (invalid argument)
[2022-06-02 17:01:11][fatal][trt_infer.cpp:340]:execute fail, code 0[cudaSuccess], message no error

looking forward your reply!!!

训练完pth转onnx后转trt？

显示错误，源码读图会自动在list添加None导致的
val = img.shape[-2:]
AttributeError: 'NoneType' object has no attribute 'shape'
咋整啊这，x02没问题