add feature/infer demo/se resnext50 #11708

tensor-tang · 2018-06-26T03:48:11Z

merged #11510 and add se-renext

… demo/mobilenet

…snext50

…-resnext50

Superjomn · 2018-06-26T03:52:09Z

paddle/contrib/inference/demo/README.md

+
+```sh
+./mobilenet_inference_demo --modeldir <model> --data <datafile>
+```


add description for resnet

Superjomn · 2018-06-26T03:52:54Z

paddle/contrib/inference/demo/CMakeLists.txt

@@ -14,3 +14,40 @@
 #

 inference_api_test(simple_on_word2vec ARGS test_word2vec)
+
+set(DEMO_INSTALL_DIR "${THIRD_PARTY_PATH}/inference_demo")
+set(mobilenet_url http://paddlemodels.bj.bcebos.com/inference-vis-demos%2Fmobilenet.tar.gz)


no need to set a variable for url, just pass it as the argument of inference_download_test_demo

luotao1 · 2018-06-26T06:34:11Z

paddle/contrib/inference/demo/vis_demo.cc

+ */
+void Main(bool use_gpu) {
+  NativeConfig config;
+  // config.model_dir = FLAGS_modeldir;


78行去掉

luotao1 · 2018-06-26T06:34:24Z

paddle/contrib/inference/demo/vis_demo.cc

+  config.fraction_of_gpu_memory = 0.1;
+#ifdef PADDLE_WITH_CUDA
+  config.fraction_of_gpu_memory = FLAGS_fraction_of_gpu_memory_to_use;
+#endif


82行需要留着么？

可以不要的。

luotao1 · 2018-06-26T06:35:06Z

paddle/contrib/inference/demo/vis_demo.cc

+  LOG(INFO) << "output: " << SummaryTensor(tensor);
+}
+
+TEST(demo, vis_demo) { Main(false /*use_gpu*/); }


需要再测下gpu的

luotao1 · 2018-06-26T06:36:50Z

paddle/contrib/inference/demo/CMakeLists.txt

@@ -14,3 +14,42 @@
 #

 inference_api_test(simple_on_word2vec ARGS test_word2vec)
+
+set(DEMO_INSTALL_DIR "${THIRD_PARTY_PATH}/inference_demo")


能否下载到build的当前路径，不要放到thirdparty路径下

直接放到build路径？还是放到别的什么路径 @Superjomn @luotao1

…-resnext50

tensor-tang · 2018-06-26T07:55:14Z

https://paddleci.ngrok.io/viewLog.html?tab=buildLog&logTab=tree&filter=debug&expand=all&buildId=85&_focus=3038

[Step 1/1] /paddle/paddle/fluid/operators/distributed/rpc_client.h:61:26: error: 'FLAGS_grpc_deadline' was not declared in this scope
[07:40:02][Step 1/1]        int64_t time_out = FLAGS_grpc_deadline) = 0;
[07:40:02][Step 1/1]                           ^
[07:40:02][Step 1/1] make[2]: *** [paddle/fluid/operators/distributed/CMakeFiles/sendrecvop_grpc.dir/rpc_client.cc.o] Error 1
[07:40:02][Step 1/1] make[2]: *** Waiting for unfinished jobs....
[07:40:02][Step 1/1] paddle/fluid/operators/distributed/CMakeFiles/sendrecvop_grpc.dir/build.make:158: recipe for target 'paddle/fluid/operators/distributed/CMakeFiles/sendrecvop_grpc.dir/rpc_client.cc.o' failed
[07:40:04][Step 1/1] In file included from /paddle/paddle/fluid/operators/distributed/grpc_client.h:41:0,
[07:40:04][Step 1/1]                  from /paddle/paddle/fluid/operators/distributed/grpc_client.cc:15:
[07:40:04][Step 1/1] /paddle/paddle/fluid/operators/distributed/rpc_client.h:61:26: error: 'FLAGS_grpc_deadline' was not declared in this scope
[07:40:04][Step 1/1]        int64_t time_out = FLAGS_grpc_deadline) = 0;
[07:40:04][Step 1/1]                           ^
[07:40:04][Step 1/1] In file included from /paddle/paddle/fluid/operators/distributed/grpc_client.cc:15:0:
[07:40:04][Step 1/1] /paddle/paddle/fluid/operators/distributed/grpc_client.h:215:49: error: 'FLAGS_grpc_deadline' was not declared in this scope
[07:40:04][Step 1/1]                               int64_t time_out = FLAGS_grpc_deadline) override;

luotao1 · 2018-06-26T07:59:54Z

paddle/contrib/inference/demo/vis_demo.cc

+    case PaddleDType::FLOAT32:
+      for (size_t i = 0; i < numel; ++i) {
+        EXPECT_NEAR(
+            static_cast<float*>(output.data.data())[i], refer.data[i], 1e-3);


1e-3太大了。改成1e-7吧。

1e-7本地gpu跑了下，太小了。改为1e-6了。

…-resnext50

tensor-tang · 2018-06-26T12:45:13Z

Failed log:

[Step 1/1] The following tests FAILED:
[10:01:36][Step 1/1] 	351 - mobilenet_inference_demo (OTHER_FAULT)
[10:01:36][Step 1/1] 	353 - ocr_inference_demo (Failed)

ocr cpu check failed

626 10:01:33.921309 50423 vis_demo.cc:68] data size 36912
[10:01:36][Step 1/1] I0626 10:01:33.921319 50423 vis_demo.cc:69] data shape size 4
[10:01:36][Step 1/1] I0626 10:01:33.921378 50423 vis_demo.cc:132] run executor
[10:01:36][Step 1/1] I0626 10:01:34.142024 50423 vis_demo.cc:136] output.size 1
[10:01:36][Step 1/1] I0626 10:01:34.142166 50423 vis_demo.cc:138] output: data[:10]	5.27365e-08 3.29674e-07 1.87326e-08 3.40372e-08 3.38332e-08 2.02055e-08 3.76429e-06 2.42356e-06 1.84867e-07 2.43073e-08 
[10:01:36][Step 1/1] I0626 10:01:34.142668 50423 vis_demo.cc:50] process a line
[10:01:36][Step 1/1] I0626 10:01:34.150573 50423 vis_demo.cc:68] data size 9312
[10:01:36][Step 1/1] I0626 10:01:34.150604 50423 vis_demo.cc:69] data shape size 3
[10:01:36][Step 1/1] I0626 10:01:34.151135 50423 vis_demo.cc:81] predictor output numel 9312
[10:01:36][Step 1/1] I0626 10:01:34.151161 50423 vis_demo.cc:82] reference output numel 9312
[10:01:36][Step 1/1] /paddle/paddle/contrib/inference/demo/vis_demo.cc:94: Failure
[10:01:36][Step 1/1] Errors while running CTest
[10:01:36][Step 1/1] The difference between static_cast<float*>(output.data.data())[i] and refer.data[i] is 5.3048133850097656e-06, which exceeds 1e-6, where
[10:01:36][Step 1/1] static_cast<float*>(output.data.data())[i] evaluates to 0.47985932230949402,
[10:01:36][Step 1/1] refer.data[i] evaluates to 0.47985401749610901, and
[10:01:36][Step 1/1] 1e-6 evaluates to 9.9999999999999995e-07.
[10:01:36][Step 1/1] /paddle/paddle/contrib/inference/demo/vis_demo.cc:94: Failure
...

mobilenet cpu data size check error:

[Step 1/1] 351/353 Test #351: mobilenet_inference_demo ........................***Exception: Other  8.56 sec
[10:01:33][Step 1/1] [==========] Running 2 tests from 1 test case.
[10:01:33][Step 1/1] [----------] Global test environment set-up.
[10:01:33][Step 1/1] [----------] 2 tests from demo
[10:01:33][Step 1/1] [ RUN      ] demo.vis_demo_cpu
[10:01:33][Step 1/1] WARNING: Logging before InitGoogleLogging() is written to STDERR
[10:01:33][Step 1/1] I0626 10:01:31.829222 50410 vis_demo.cc:113] init predictor
[10:01:33][Step 1/1] I0626 10:01:31.846272 50410 vis_demo.cc:117] begin to process data
[10:01:33][Step 1/1] I0626 10:01:31.853760 50410 vis_demo.cc:50] process a line
[10:01:33][Step 1/1] I0626 10:01:31.913990 50410 vis_demo.cc:68] data size 270000
[10:01:33][Step 1/1] I0626 10:01:31.914029 50410 vis_demo.cc:69] data shape size 4
[10:01:33][Step 1/1] I0626 10:01:31.915087 50410 vis_demo.cc:132] run executor
[10:01:33][Step 1/1] I0626 10:01:32.192790 50410 vis_demo.cc:136] output.size 1
[10:01:33][Step 1/1] I0626 10:01:32.192885 50410 vis_demo.cc:138] output: data[:10]	3 0.0267651 0.595492 0.553865 0.710751 0.837597 5 0.0162648 0.303328 -0.00768402 
[10:01:33][Step 1/1] I0626 10:01:32.192948 50410 vis_demo.cc:50] process a line
[10:01:33][Step 1/1] F0626 10:01:32.192968 50410 vis_demo.cc:53] Check failed: columns.size() == 2UL (0 vs. 2) data format error, should be <data>	<shape>
[10:01:33][Step 1/1] *** Check failure stack trace: ***

tensor-tang · 2018-06-26T15:52:39Z

Still have this issue:

351/353 Test #351: mobilenet_inference_demo ........................***Exception: Other  7.91 sec
[15:20:13][Step 1/1] [==========] Running 2 tests from 1 test case.
[15:20:13][Step 1/1] [----------] Global test environment set-up.
[15:20:13][Step 1/1] [----------] 2 tests from demo
[15:20:13][Step 1/1] [ RUN      ] demo.vis_demo_cpu
[15:20:13][Step 1/1] WARNING: Logging before InitGoogleLogging() is written to STDERR
[15:20:13][Step 1/1] I0626 15:20:11.055225 48676 vis_demo.cc:113] init predictor
[15:20:13][Step 1/1] I0626 15:20:11.104532 48676 vis_demo.cc:117] begin to process data
[15:20:13][Step 1/1] I0626 15:20:11.117425 48676 vis_demo.cc:50] process a line
[15:20:13][Step 1/1] I0626 15:20:11.245939 48676 vis_demo.cc:68] data size 270000
[15:20:13][Step 1/1] I0626 15:20:11.245976 48676 vis_demo.cc:69] data shape size 4
[15:20:13][Step 1/1] I0626 15:20:11.247333 48676 vis_demo.cc:132] run executor
[15:20:13][Step 1/1] I0626 15:20:11.545816 48676 vis_demo.cc:136] output.size 1
[15:20:13][Step 1/1] I0626 15:20:11.545897 48676 vis_demo.cc:138] output: data[:10]	3 0.0267651 0.595492 0.553865 0.710751 0.837597 5 0.0162648 0.303328 -0.00768402 
[15:20:13][Step 1/1] I0626 15:20:11.545958 48676 vis_demo.cc:50] process a line
[15:20:13][Step 1/1] F0626 15:20:11.545975 48676 vis_demo.cc:53] Check failed: columns.size() == 2UL (0 vs. 2) data format error, should be <data>	<shape>
[15:20:13][Step 1/1] *** Check failure stack trace: ***

But it can pass on my local machine.

@luotao1 Could you help double check on your local machine?

BTW the data has been download, https://paddleci.ngrok.io/viewLog.html?tab=buildLog&logTab=tree&filter=debug&expand=all&buildId=164&_focus=2973

[Step 1/1] -- inference demo /paddle/build/inference_demo/mobilenet_inference_demo
[14:46:07][Step 1/1] -- Download mobilenet_inference_demo model from http://paddlemodels.bj.bcebos.com/inference-vis-demos%2Fmobilenet.tar.gz

…-resnext50

luotao1 · 2018-06-27T02:38:48Z

Could you help double check on your local machine

@tensor-tang I am OK on my local machine. You can add md5sum to check the success of Download.

luotao1

LGTM

Superjomn and others added 12 commits June 15, 2018 06:34

Init

22aafcb

init

f5d0783

add automatic model deploy

2e739cf

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

190a871

… demo/mobilenet

upadte

904297e

update

1c0bbfd

update

c34da73

refine

8b827d0

Merge branch 'Superjomn-demo/mobilenet' into feature/infer_demo/se-re…

c2d9c42

…snext50

add se-renext50 infer demo

c583513

update url

c94cacc

Merge remote-tracking branch 'ups/develop' into feature/infer_demo/se…

75e5084

…-resnext50

tensor-tang requested review from Superjomn and luotao1 June 26, 2018 03:48

Superjomn assigned Superjomn and tensor-tang Jun 26, 2018

Superjomn reviewed Jun 26, 2018

View reviewed changes

tensor-tang added 2 commits June 26, 2018 14:22

refine and add ocr

fc568f4

update readme

e8b0dcb

luotao1 reviewed Jun 26, 2018

View reviewed changes

tensor-tang added 3 commits June 26, 2018 15:09

add check refer

d509021

update and enable gpu test

7449fe7

Merge remote-tracking branch 'ups/develop' into feature/infer_demo/se…

46285ed

…-resnext50

luotao1 reviewed Jun 26, 2018

View reviewed changes

tensor-tang added 2 commits June 26, 2018 16:18

fix issue on gpu

5c8e323

Merge remote-tracking branch 'ups/develop' into feature/infer_demo/se…

7e7df49

…-resnext50

update threshold

33b8ea8

tensor-tang force-pushed the feature/infer_demo/se-resnext50 branch from 045b77c to 33b8ea8 Compare June 26, 2018 12:57

tensor-tang added 2 commits June 27, 2018 10:13

disable inference mobilenet test

ba0401a

Merge remote-tracking branch 'ups/develop' into feature/infer_demo/se…

529915c

…-resnext50

luotao1 approved these changes Jun 27, 2018

View reviewed changes

Superjomn assigned luotao1 Jun 27, 2018

Superjomn merged commit 5299387 into PaddlePaddle:develop Jun 27, 2018

tensor-tang deleted the feature/infer_demo/se-resnext50 branch June 27, 2018 03:13

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

add feature/infer demo/se resnext50 #11708

add feature/infer demo/se resnext50 #11708

tensor-tang commented Jun 26, 2018 •

edited

Loading

Superjomn Jun 26, 2018

tensor-tang Jun 26, 2018

tensor-tang Jun 26, 2018

Superjomn Jun 26, 2018

tensor-tang Jun 26, 2018

luotao1 Jun 26, 2018

tensor-tang Jun 26, 2018

luotao1 Jun 26, 2018

tensor-tang Jun 26, 2018

luotao1 Jun 26, 2018

luotao1 Jun 26, 2018

tensor-tang Jun 26, 2018

tensor-tang commented Jun 26, 2018

luotao1 Jun 26, 2018

tensor-tang Jun 26, 2018

tensor-tang commented Jun 26, 2018

tensor-tang commented Jun 26, 2018 •

edited

Loading

luotao1 commented Jun 27, 2018

luotao1 left a comment

add feature/infer demo/se resnext50 #11708

add feature/infer demo/se resnext50 #11708

Conversation

tensor-tang commented Jun 26, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tensor-tang commented Jun 26, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tensor-tang commented Jun 26, 2018

tensor-tang commented Jun 26, 2018 • edited Loading

luotao1 commented Jun 27, 2018

luotao1 left a comment

Choose a reason for hiding this comment

tensor-tang commented Jun 26, 2018 •

edited

Loading

tensor-tang commented Jun 26, 2018 •

edited

Loading