search

layerism / TensorRT-Inference-Server-Tutorial

服务侧深度学习部署案例
454 stars 73 forks source link

这个项目很棒！但是遇到了一些问题，求助大神仓主，我把环境都配置好了，运行client.sh报错．．． #12

Open Byronnar opened 4 years ago

Byronnar commented 4 years ago

''' I0511 10:04:23.546782 1 model_repository_manager.cc:807] successfully loaded 'detection' version 1 I0511 10:04:23.546846 1 main.cc:455] Starting endpoints, 'inference:0' listening on I0511 10:04:23.550919 1 grpc_server.cc:1800] Started GRPCService at 0.0.0.0:8001 I0511 10:04:23.550956 1 http_server.cc:1205] Starting HTTPService at 0.0.0.0:8000 I0511 10:04:23.593179 1 http_server.cc:1220] Starting Metrics Service at 0.0.0.0:8002 2020-05-11 10:04:46.130023: E tensorflow/core/common_runtime/executor.cc:648] Executor failed to create kernel. Invalid argument: NodeDef mentions attr 'incompatible_shape_error' not in Op<name=NotEqual; signature=x:T, y:T -> z:bool; attr=T:type,allowed=[DT_BFLOAT16, DT_HALF, DT_FLOAT, DT_DOUBLE, DT_UINT8, ..., DT_QINT8, DT_QINT32, DT_STRING, DT_BOOL, DT_COMPLEX128]; is_commutative=true>; NodeDef: {{node image_decoder/decode_image/cond_jpeg/cond_png/cond_gif/check_gif_channels_1}}. (Check whether your GraphDef-interpreting binary is up to date with your GraphDef-generating binary.). [[image_decoder/decode_image/cond_jpeg/cond_png/cond_gif/check_gif_channels_1]] 2020-05-11 10:04:46.150321: E tensorflow/core/common_runtime/executor.cc:648] Executor failed to create kernel. Invalid argument: NodeDef mentions attr 'incompatible_shape_error' not in Op<name=NotEqual; signature=x:T, y:T -> z:bool; attr=T:type,allowed=[DT_BFLOAT16, DT_HALF, DT_FLOAT, DT_DOUBLE, DT_UINT8, ..., DT_QINT8, DT_QINT32, DT_STRING, DT_BOOL, DT_COMPLEX128]; is_commutative=true>; NodeDef: {{node image_decoder/decode_image/cond_jpeg/check_jpeg_channels}}. (Check whether your GraphDef-interpreting binary is up to date with your GraphDef-generating binary.). [[image_decoder/decode_image/cond_jpeg/check_jpeg_channels]]

''' 请问是什么原因呢？希望能回复一下．各路大神看到希望也能帮帮忙

Byronnar commented 4 years ago

查了一下，说是训练模型的版本跟部署版本不一致，但是您是用pytorch训练的吧， This could be caused by different tensorflow versions used to train and serve the model.

Byronnar commented 4 years ago

@layerism

17702513221 commented 4 years ago

作者的模型版本不是官方的，是pytorch复现的

WangYaohui6233 commented 4 years ago

我的理解，这个是前处理的部分报的错吧，和模型没关系吧，我用tf.io.decode_image这个也报了上面的错，我换了tf.io.decode_jpeg这个接口，另外的错： image

我的客户端是这样写的： image

服务端的前处理是这样写的： image

WangYaohui6233 commented 4 years ago

我单独写了个测试程序，上面写的客户端和服务端的代码应该是没问题的： image

Byronnar commented 4 years ago

所以这个有人完全跑通了的吗

nobody-cheng commented 4 years ago

所以这个有人完全跑通了的吗

这里遇到大佬，请问是不是有准备要把yolov3用这个来搭建