基於樹莓派4B與Paddle-Lite實現的實時口罩識別

時間 2020-04-09

標籤基於樹莓 4b paddle lite 實現實時口罩識別简体版

原文原文鏈接

文檔結構

0 項目簡介	1 環境搭建	2 快速開始	3 關於模型	4 關於opt模型轉化工具	5 FAQ
	1.1硬件環境		3.1模型更新
--------	--------	--------	--------	--------	--------
	1.2軟件環境
--------	--------	--------	--------	--------	--------
	1.3運行環境

0 項目簡介

此項目仿照Paddle-Lite-Demo格式，將Android的C++口罩識別demo移植到樹莓派上，並增長了實時識別的功能。html

源碼node

已經做爲數據集的形式存放在data/data24081/mask.zip。python

————————————————————————————————————————————————————————————————————————————————————linux

經過此項目，你能夠學習到：git

1.實如今樹莓派上或者像RK3399這種開發板上實現對**圖片**或者**使用攝像頭**進行實時識別

2.opt模型轉化工具的使用

3.PaddleHub下載模型的方法

此項目使用armv8的模型github

關於armv7hf，或者使用32位系統的方法我寫到了最下面~web

先展現一下最後的效果，用手遮擋了一下仍是能夠識別的出的，效果還不錯～api

下載安裝命令

## CPU版本安裝命令
pip install -f https://paddlepaddle.org.cn/pip/oschina/cpu paddlepaddle

## GPU版本安裝命令
pip install -f https://paddlepaddle.org.cn/pip/oschina/gpu paddlepaddle-gpu

1 環境搭建

1.1硬件環境

開發板：樹莓派4B架構

CPU：博通BCM2711，4核A72的CPU，armv8架構處理器內存：LPDDR4 4GBapp

攝像頭：（CSI接口攝像頭500萬像素+15cm軟排線，某寶不到20塊錢就能買的到）若是有UVC免驅動的USB攝像頭的話也是能夠的，注意分辨率不要超過720p。

1.2軟件環境

使用的操做系統是Debian-Pi-Aarch64，附上連接：https://gitee.com/openfans-community/Debian-Pi-Aarch64/

（armv8處理器配合64位系統才能發揮出更強的性能啦～）

這個系統系統是64位的，作的很精緻，速度比官方32位快不少，也比Ubuntu18.04sever64位環境容易搭建不少，推薦使用！

若是你使用的是這個系統的話，使用CSI攝像頭須要在/boot/config.txt文件中將start_x=1的註釋去掉，重啓後ls /dev/video*能看到有/dev/video0的話，CSI攝像頭就能夠正常使用了（後面的video10 video11 video12不是CSI攝像頭）

1.3運行環境

參照Demo咱們須要安裝gcc g++ make wget unzip libopencv-dev pkg-config和CMake

$ sudo apt-get update
$ sudo apt-get install gcc g++ make wget unzip libopencv-dev pkg-config
$ wget https://www.cmake.org/files/v3.10/cmake-3.10.3.tar.gz
$ tar -zxvf cmake-3.10.3.tar.gz
$ cd cmake-3.10.3
$ ./configure && make -j4 && sudo make install

注意在安裝libopencv-dev或者g++的時候可能會報依賴錯誤，這個時候須要使用aptitude來解決依賴問題

sudo apt install aptitude

sudo aptitude install libopencv-dev

而後通常提供的第一個解決方案不會作任何改變，因此按N會從新推薦一套方案。
第二套方案通常是把這些高版本的包降級處理來解決依賴問題，按Y確認等待安裝完畢便可。

項目使用的Paddle-lite是我本身編譯好的，版本是v2.3.0，若是想替換成本身編譯好的庫能夠在mask_detection_demo/Paddle-Lite下自行替換include和libs

2 快速開始

若是前面的準備工做都已經完畢的話，下載源碼data/data24129/mask.zip。移動到樹莓派4B中解壓,下面是對項目代碼結構簡單的說明

mask_detection
  		Paddle-Lite：	
				include (編譯好的Paddle—Lite的頭文件)
				libs（編譯好的Paddle—Lite的庫文件）
  		code：
				model(模型連接(https://paddle-inference-dist.bj.bcebos.com/mask_detection.tar.gz))
				images(測試圖片)
				CMakeLists.txt
				mask_detection.cc
				run.sh(無參數時使用攝像頭實時識別，不然須要跟定測試圖片的路徑)

此項目可對圖片或實時的視頻流進行口罩檢測，進入code/文件夾在終端裏運行run.sh

若是使用攝像頭進行實時檢測直接運行，無需跟參數（按「q」退出）
./run.sh
若對圖片進行檢測，需添加圖片的路徑（按「0」退出）
./run.sh ../images/test_mask_detection.jpg

效果以下：

圖片識別

實時視頻流識別

3 關於模型

當前使用的模型仍是v2.2.0以前的模型,若想更換最新版的模型請從PaddleHub上進行下載～

下載安裝命令

## CPU版本安裝命令
pip install -f https://paddlepaddle.org.cn/pip/oschina/cpu paddlepaddle

## GPU版本安裝命令
pip install -f https://paddlepaddle.org.cn/pip/oschina/gpu paddlepaddle-gpu

項目裏的模型有兩個，一個是對口罩識別的模型，一個是對人臉識別的模型。基本的邏輯就是先檢測人臉，檢測到後再判斷是否佩戴口罩。

3.1模型更新

最新的代碼經過PaddleHub下載

目前口罩識別的模型最新版是１.2.0

pip安裝PaddleHub(個人PC系統是Ubuntu18.04.4LTS)

pip3 install paddlehub

而後安裝最新版模型

hub install pyramidbox_lite_mobile==1.1.0
hub install pyramidbox_lite_mobile_mask==1.2.0

下載最新模型：

import paddlehub as hub
pyramidbox_lite_mobile_mask = hub.Module(name="pyramidbox_lite_mobile_mask")
# 將模型保存在test_program文件夾之中
pyramidbox_lite_mobile_mask.processor.save_inference_model(dirname="test_program")

下載好後使用opt模型轉化工具

轉化命令:

opt --model_file=./__model__ --param_file=./__params__ --optimize_out_type=naive_buffer --optimize_out=model

4 關於opt模型轉化工具

opt工具是Paddle-Lite的模型轉化工具，這裏使用它的主要目的是將PaddlePaddle上訓練好的模型轉化成Lite可用來預測的模型

關於opt工具的官方文檔以下：opt

注意opt工具是x86下使用的，是不可以在arm開發板上使用的

因此咱們須要一個x86下的Linux環境，這裏咱們能夠藉助AiStudio來實現opt模型轉化

這裏給你們感覺一下opt工具的使用方法：

In[1]

#先下載opt模型轉化工具，這裏咱們使用官方編譯好的，想本身編譯的可根據官方的教程，連接以下：[教程](https://paddle-lite.readthedocs.io/zh/latest/user_guides/model_optimize_tool.html)
!wget https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.3.0/opt

--2020-03-12 12:39:29--  https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.3.0/opt
Resolving github.com (github.com)... 13.250.177.223
Connecting to github.com (github.com)|13.250.177.223|:443... connected.
HTTP request sent, awaiting response... 302 Found
Location: https://github-production-release-asset-2e65be.s3.amazonaws.com/104208128/1127c300-5883-11ea-8d46-3c94eb9ccdc3?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAIWNJYAX4CSVEH53A%2F20200312%2Fus-east-1%2Fs3%2Faws4_request&X-Amz-Date=20200312T043930Z&X-Amz-Expires=300&X-Amz-Signature=5bc29c36021ef6acc01b9b86a0aa6ce81802132c514b9333d70b4f1ad2f7f178&X-Amz-SignedHeaders=host&actor_id=0&response-content-disposition=attachment%3B%20filename%3Dopt&response-content-type=application%2Foctet-stream [following]
--2020-03-12 12:39:30--  https://github-production-release-asset-2e65be.s3.amazonaws.com/104208128/1127c300-5883-11ea-8d46-3c94eb9ccdc3?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAIWNJYAX4CSVEH53A%2F20200312%2Fus-east-1%2Fs3%2Faws4_request&X-Amz-Date=20200312T043930Z&X-Amz-Expires=300&X-Amz-Signature=5bc29c36021ef6acc01b9b86a0aa6ce81802132c514b9333d70b4f1ad2f7f178&X-Amz-SignedHeaders=host&actor_id=0&response-content-disposition=attachment%3B%20filename%3Dopt&response-content-type=application%2Foctet-stream
Resolving github-production-release-asset-2e65be.s3.amazonaws.com (github-production-release-asset-2e65be.s3.amazonaws.com)... 52.216.96.99
Connecting to github-production-release-asset-2e65be.s3.amazonaws.com (github-production-release-asset-2e65be.s3.amazonaws.com)|52.216.96.99|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 13423408 (13M) [application/octet-stream]
Saving to: ‘opt’

opt                  21%[===>                ]   2.77M  1.07KB/s    eta 28m 48s^C

而後下載須要轉化的模型

In[10]

#使用hub下載命令下載模型
!hub install pyramidbox_lite_mobile==1.1.0
!hub install pyramidbox_lite_mobile_mask==1.2.0

#將模型轉化爲Combined形式的模型，即__model__ 和 __params__
import paddlehub as hub
pyramidbox_lite_mobile_mask = hub.Module(name="pyramidbox_lite_mobile_mask")
# 將模型保存在test_program文件夾之中
pyramidbox_lite_mobile_mask.processor.save_inference_model(dirname="test_program") 

%cd ~

#給opt工具加上可執行權限
!chmod +x opt
#將opt工具複製到兩個模型的文件夾下
!cp opt test_program/mask_detector/ && cp opt test_program/pyramidbox_lite
#進入到口罩模型文件夾
%cd test_program/mask_detector/
#使用opt工具進行模型轉化 將__model__ 和 __params__ 轉化爲model.nb 
!./opt --model_file=./__model__ --param_file=./__params__ --optimize_out_type=naive_buffer --optimize_out=model
!ls

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/sklearn/externals/joblib/externals/cloudpickle/cloudpickle.py:47: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses
  import imp
Module pyramidbox_lite_mobile-1.1.0 already installed in /home/aistudio/.paddlehub/modules/pyramidbox_lite_mobile
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/sklearn/externals/joblib/externals/cloudpickle/cloudpickle.py:47: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses
  import imp
Module pyramidbox_lite_mobile_mask-1.2.0 already installed in /home/aistudio/.paddlehub/modules/pyramidbox_lite_mobile_mask

[2020-03-12 13:11:01,130] [    INFO] - Installing pyramidbox_lite_mobile_mask module
[2020-03-12 13:11:01,146] [    INFO] - Module pyramidbox_lite_mobile_mask already installed in /home/aistudio/.paddlehub/modules/pyramidbox_lite_mobile_mask
[2020-03-12 13:11:01,237] [    INFO] - Installing pyramidbox_lite_mobile module
[2020-03-12 13:11:01,250] [    INFO] - Module pyramidbox_lite_mobile already installed in /home/aistudio/.paddlehub/modules/pyramidbox_lite_mobile

/home/aistudio
/home/aistudio/test_program/mask_detector
[W  3/12 13:11: 2.441 ...dle/hongming/Paddle-Lite/lite/api/opt.cc:138 RunOptimize] Load combined-param model. Option model_dir will be ignored
[I  3/12 13:11: 2.441 ...hongming/Paddle-Lite/lite/api/cxx_api.cc:244 Build] Load model from file.
[I  3/12 13:11: 2.451 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_quant_dequant_fuse_pass
[I  3/12 13:11: 2.454 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_quant_dequant_fuse_pass
[I  3/12 13:11: 2.454 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: weight_quantization_preprocess_pass
[I  3/12 13:11: 2.454 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: weight_quantization_preprocess_pass
[I  3/12 13:11: 2.454 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_conv_elementwise_fuse_pass
[I  3/12 13:11: 2.457 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 21 subgraph
[I  3/12 13:11: 2.458 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_conv_elementwise_fuse_pass
[I  3/12 13:11: 2.458 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_conv_bn_fuse_pass
[I  3/12 13:11: 2.465 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 21 subgraph
[I  3/12 13:11: 2.467 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 1 subgraph
[I  3/12 13:11: 2.467 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_conv_bn_fuse_pass
[I  3/12 13:11: 2.467 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_conv_elementwise_fuse_pass
[I  3/12 13:11: 2.469 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_conv_elementwise_fuse_pass
[I  3/12 13:11: 2.469 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_conv_activation_fuse_pass
[I  3/12 13:11: 2.470 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 14 subgraph
[I  3/12 13:11: 2.474 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_conv_activation_fuse_pass
[I  3/12 13:11: 2.474 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_var_conv_2d_activation_fuse_pass
[I  3/12 13:11: 2.474 ...ngming/Paddle-Lite/lite/core/optimizer.h:177 RunPasses]    - Skip lite_var_conv_2d_activation_fuse_pass because the target or kernel does not match.
[I  3/12 13:11: 2.474 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_fc_fuse_pass
[I  3/12 13:11: 2.474 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 1 subgraph
[I  3/12 13:11: 2.474 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_fc_fuse_pass
[I  3/12 13:11: 2.474 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_shuffle_channel_fuse_pass
[I  3/12 13:11: 2.474 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_shuffle_channel_fuse_pass
[I  3/12 13:11: 2.474 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_transpose_softmax_transpose_fuse_pass
[I  3/12 13:11: 2.474 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_transpose_softmax_transpose_fuse_pass
[I  3/12 13:11: 2.474 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_interpolate_fuse_pass
[I  3/12 13:11: 2.475 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_interpolate_fuse_pass
[I  3/12 13:11: 2.475 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: identity_scale_eliminate_pass
[I  3/12 13:11: 2.475 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 1 subgraph
[I  3/12 13:11: 2.475 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: identity_scale_eliminate_pass
[I  3/12 13:11: 2.475 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: elementwise_mul_constant_eliminate_pass
[I  3/12 13:11: 2.475 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: elementwise_mul_constant_eliminate_pass
[I  3/12 13:11: 2.475 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_sequence_pool_concat_fuse_pass
[I  3/12 13:11: 2.475 ...ngming/Paddle-Lite/lite/core/optimizer.h:177 RunPasses]    - Skip lite_sequence_pool_concat_fuse_pass because the target or kernel does not match.
[I  3/12 13:11: 2.475 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: static_kernel_pick_pass
[I  3/12 13:11: 2.476 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: static_kernel_pick_pass
[I  3/12 13:11: 2.476 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: variable_place_inference_pass
[I  3/12 13:11: 2.478 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: variable_place_inference_pass
[I  3/12 13:11: 2.478 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: argument_type_display_pass
[I  3/12 13:11: 2.478 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: argument_type_display_pass
[I  3/12 13:11: 2.478 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: type_target_cast_pass
[I  3/12 13:11: 2.478 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: type_target_cast_pass
[I  3/12 13:11: 2.478 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: variable_place_inference_pass
[I  3/12 13:11: 2.480 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: variable_place_inference_pass
[I  3/12 13:11: 2.480 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: argument_type_display_pass
[I  3/12 13:11: 2.480 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: argument_type_display_pass
[I  3/12 13:11: 2.480 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: io_copy_kernel_pick_pass
[I  3/12 13:11: 2.480 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: io_copy_kernel_pick_pass
[I  3/12 13:11: 2.480 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: argument_type_display_pass
[I  3/12 13:11: 2.480 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: argument_type_display_pass
[I  3/12 13:11: 2.480 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: variable_place_inference_pass
[I  3/12 13:11: 2.481 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: variable_place_inference_pass
[I  3/12 13:11: 2.481 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: argument_type_display_pass
[I  3/12 13:11: 2.481 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: argument_type_display_pass
[I  3/12 13:11: 2.481 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: type_precision_cast_pass
[I  3/12 13:11: 2.482 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: type_precision_cast_pass
[I  3/12 13:11: 2.482 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: variable_place_inference_pass
[I  3/12 13:11: 2.483 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: variable_place_inference_pass
[I  3/12 13:11: 2.483 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: argument_type_display_pass
[I  3/12 13:11: 2.483 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: argument_type_display_pass
[I  3/12 13:11: 2.483 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: type_layout_cast_pass
dot:
digraph G {
   node_108[label="@HUB_pyramidbox_lite_mobile_mask@fc1_glass.b_0"]
   node_107[label="@HUB_pyramidbox_lite_mobile_mask@fc1_glass.w_0"]
   node_106[label="fc30" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_104[label="@HUB_pyramidbox_lite_mobile_mask@age_block3_2.w_0"]
   node_102[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn5_1.b_0"]
   node_101[label="@HUB_pyramidbox_lite_mobile_mask@age_block5_1.w_0"]
   node_99[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn1_1.b_0"]
   node_103[label="conv2d29" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_98[label="@HUB_pyramidbox_lite_mobile_mask@age_block1_1.w_0"]
   node_97[label="conv2d27" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_96[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn2_2.b_0"]
   node_29[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_9.tmp_2"]
   node_13[label="conv2d5" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_38[label="@HUB_pyramidbox_lite_mobile_mask@age_block5_3.w_0"]
   node_36[label="@HUB_pyramidbox_lite_mobile_mask@tmp_1"]
   node_34[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_12.tmp_2"]
   node_61[label="conv2d17" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_31[label="@HUB_pyramidbox_lite_mobile_mask@age_block4_3.w_0"]
   node_26[label="@HUB_pyramidbox_lite_mobile_mask@age_block3_3.w_0"]
   node_39[label="@HUB_pyramidbox_lite_mobile_mask@relu_10.tmp_0"]
   node_25[label="conv2d8" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_59[label="@HUB_pyramidbox_lite_mobile_mask@glass_block5_2.w_0"]
   node_42[label="conv2d12" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_27[label="@HUB_pyramidbox_lite_mobile_mask@relu_6.tmp_0"]
   node_24[label="@HUB_pyramidbox_lite_mobile_mask@tmp_0"]
   node_40[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn5_3.b_0"]
   node_9[label="concat3" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_32[label="@HUB_pyramidbox_lite_mobile_mask@relu_8.tmp_0"]
   node_92[label="@HUB_pyramidbox_lite_mobile_mask@age_block1_2.w_0"]
   node_4[label="@HUB_pyramidbox_lite_mobile_mask@crelu.w_0"]
   node_10[label="@HUB_pyramidbox_lite_mobile_mask@concat_0.tmp_0"]
   node_18[label="conv2d6" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_84[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn2_1.b_0"]
   node_94[label="conv2d26" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_43[label="@HUB_pyramidbox_lite_mobile_mask@glass_block4_3.w_0"]
   node_30[label="conv2d9" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_11[label="relu4" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_15[label="@HUB_pyramidbox_lite_mobile_mask@relu_2.tmp_0"]
   node_35[label="elementwise_add10" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_71[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn5_2.b_0"]
   node_7[label="scale2" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_47[label="conv2d13" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_78[label="@HUB_pyramidbox_lite_mobile_mask@relu_7.tmp_0"]
   node_85[label="@HUB_pyramidbox_lite_mobile_mask@relu_3.tmp_0"]
   node_5[label="@HUB_pyramidbox_lite_mobile_mask@crelu_bn.b_0"]
   node_2[label="@HUB_pyramidbox_lite_mobile_mask@image"]
   node_28[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn3_3.b_0"]
   node_3[label="conv2d1" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_66[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn4_1.b_0"]
   node_6[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_0.tmp_2"]
   node_33[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn4_3.b_0"]
   node_51[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2"]
   node_44[label="@HUB_pyramidbox_lite_mobile_mask@relu_12.tmp_0"]
   node_14[label="@HUB_pyramidbox_lite_mobile_mask@age_block1_3.w_0"]
   node_1[label="feed"]
   node_89[label="@HUB_pyramidbox_lite_mobile_mask@relu_5.tmp_0"]
   node_8[label="@HUB_pyramidbox_lite_mobile_mask@scale_0.tmp_0"]
   node_20[label="@HUB_pyramidbox_lite_mobile_mask@relu_4.tmp_0"]
   node_21[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn2_3.b_0"]
   node_65[label="@HUB_pyramidbox_lite_mobile_mask@glass_block4_1.w_0"]
   node_23[label="elementwise_add7" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_16[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn1_3.b_0"]
   node_87[label="@HUB_pyramidbox_lite_mobile_mask@age_block3_1.w_0"]
   node_17[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_3.tmp_2"]
   node_19[label="@HUB_pyramidbox_lite_mobile_mask@age_block2_3.w_0"]
   node_46[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_18.tmp_2"]
   node_41[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_15.tmp_2"]
   node_79[label="conv2d22" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_81[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn4_2.b_0"]
   node_48[label="@HUB_pyramidbox_lite_mobile_mask@glass_block5_3.w_0"]
   node_100[label="conv2d28" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_62[label="@HUB_pyramidbox_lite_mobile_mask@glass_block5_1.w_0"]
   node_82[label="conv2d23" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_49[label="@HUB_pyramidbox_lite_mobile_mask@relu_14.tmp_0"]
   node_73[label="@HUB_pyramidbox_lite_mobile_mask@glass_block4_2.w_0"]
   node_53[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1"]
   node_54[label="@HUB_pyramidbox_lite_mobile_mask@save_infer_model/scale_0"]
   node_55[label="fetch15" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_56[label="fetch"]
   node_45[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn4_3.b_0"]
   node_52[label="softmax14" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_57[label="conv2d16" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_58[label="@HUB_pyramidbox_lite_mobile_mask@relu_13.tmp_0"]
   node_60[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn5_2.b_0"]
   node_12[label="@HUB_pyramidbox_lite_mobile_mask@relu_0.tmp_0"]
   node_63[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn5_1.b_0"]
   node_64[label="conv2d18" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_68[label="conv2d19" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_80[label="@HUB_pyramidbox_lite_mobile_mask@age_block4_2.w_0"]
   node_22[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_6.tmp_2"]
   node_70[label="@HUB_pyramidbox_lite_mobile_mask@age_block5_2.w_0"]
   node_50[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn5_3.b_0"]
   node_67[label="@HUB_pyramidbox_lite_mobile_mask@relu_11.tmp_0"]
   node_72[label="conv2d20" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_37[label="conv2d11" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_74[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn4_2.b_0"]
   node_77[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn4_1.b_0"]
   node_105[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn3_2.b_0"]
   node_69[label="@HUB_pyramidbox_lite_mobile_mask@relu_9.tmp_0"]
   node_75[label="conv2d21" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_76[label="@HUB_pyramidbox_lite_mobile_mask@age_block4_1.w_0"]
   node_0[label="feed0" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_83[label="@HUB_pyramidbox_lite_mobile_mask@age_block2_1.w_0"]
   node_86[label="conv2d24" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_88[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn3_1.b_0"]
   node_90[label="conv2d25" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_91[label="@HUB_pyramidbox_lite_mobile_mask@relu_1.tmp_0"]
   node_93[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn1_2.b_0"]
   node_95[label="@HUB_pyramidbox_lite_mobile_mask@age_block2_2.w_0"]
   node_1->node_0
   node_0->node_2
   node_4->node_3
   node_2->node_3
   node_5->node_3
   node_3->node_6
   node_6->node_7
   node_7->node_8
   node_6->node_9
   node_8->node_9
   node_9->node_10
   node_10->node_11
   node_11->node_12
   node_14->node_13
   node_15->node_13
   node_16->node_13
   node_13->node_17
   node_19->node_18
   node_20->node_18
   node_21->node_18
   node_18->node_22
   node_22->node_23
   node_17->node_23
   node_23->node_24
   node_26->node_25
   node_27->node_25
   node_28->node_25
   node_25->node_29
   node_31->node_30
   node_32->node_30
   node_33->node_30
   node_30->node_34
   node_34->node_35
   node_29->node_35
   node_35->node_36
   node_38->node_37
   node_39->node_37
   node_40->node_37
   node_37->node_41
   node_43->node_42
   node_44->node_42
   node_45->node_42
   node_42->node_46
   node_48->node_47
   node_49->node_47
   node_50->node_47
   node_47->node_51
   node_53->node_52
   node_52->node_54
   node_54->node_55
   node_55->node_56
   node_58->node_57
   node_59->node_57
   node_60->node_57
   node_57->node_49
   node_46->node_61
   node_62->node_61
   node_63->node_61
   node_61->node_58
   node_41->node_64
   node_65->node_64
   node_66->node_64
   node_64->node_67
   node_69->node_68
   node_70->node_68
   node_71->node_68
   node_68->node_39
   node_67->node_72
   node_73->node_72
   node_74->node_72
   node_72->node_44
   node_29->node_75
   node_76->node_75
   node_77->node_75
   node_75->node_78
   node_78->node_79
   node_80->node_79
   node_81->node_79
   node_79->node_32
   node_17->node_82
   node_83->node_82
   node_84->node_82
   node_82->node_85
   node_24->node_86
   node_87->node_86
   node_88->node_86
   node_86->node_89
   node_91->node_90
   node_92->node_90
   node_93->node_90
   node_90->node_15
   node_85->node_94
   node_95->node_94
   node_96->node_94
   node_94->node_20
   node_12->node_97
   node_98->node_97
   node_99->node_97
   node_97->node_91
   node_36->node_100
   node_101->node_100
   node_102->node_100
   node_100->node_69
   node_89->node_103
   node_104->node_103
   node_105->node_103
   node_103->node_27
   node_107->node_106
   node_51->node_106
   node_108->node_106
   node_106->node_53
} // end G
dot:
digraph G {
   node_217[label="@HUB_pyramidbox_lite_mobile_mask@fc1_glass.b_0"]
   node_216[label="@HUB_pyramidbox_lite_mobile_mask@fc1_glass.w_0"]
   node_215[label="fc30" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_213[label="@HUB_pyramidbox_lite_mobile_mask@age_block3_2.w_0"]
   node_211[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn5_1.b_0"]
   node_210[label="@HUB_pyramidbox_lite_mobile_mask@age_block5_1.w_0"]
   node_208[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn1_1.b_0"]
   node_212[label="conv2d29" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_207[label="@HUB_pyramidbox_lite_mobile_mask@age_block1_1.w_0"]
   node_206[label="conv2d27" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_205[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn2_2.b_0"]
   node_138[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_9.tmp_2"]
   node_122[label="conv2d5" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_147[label="@HUB_pyramidbox_lite_mobile_mask@age_block5_3.w_0"]
   node_145[label="@HUB_pyramidbox_lite_mobile_mask@tmp_1"]
   node_143[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_12.tmp_2"]
   node_170[label="conv2d17" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_140[label="@HUB_pyramidbox_lite_mobile_mask@age_block4_3.w_0"]
   node_135[label="@HUB_pyramidbox_lite_mobile_mask@age_block3_3.w_0"]
   node_148[label="@HUB_pyramidbox_lite_mobile_mask@relu_10.tmp_0"]
   node_134[label="conv2d8" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_168[label="@HUB_pyramidbox_lite_mobile_mask@glass_block5_2.w_0"]
   node_151[label="conv2d12" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_136[label="@HUB_pyramidbox_lite_mobile_mask@relu_6.tmp_0"]
   node_133[label="@HUB_pyramidbox_lite_mobile_mask@tmp_0"]
   node_149[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn5_3.b_0"]
   node_118[label="concat3" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_141[label="@HUB_pyramidbox_lite_mobile_mask@relu_8.tmp_0"]
   node_201[label="@HUB_pyramidbox_lite_mobile_mask@age_block1_2.w_0"]
   node_113[label="@HUB_pyramidbox_lite_mobile_mask@crelu.w_0"]
   node_119[label="@HUB_pyramidbox_lite_mobile_mask@concat_0.tmp_0"]
   node_127[label="conv2d6" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_193[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn2_1.b_0"]
   node_203[label="conv2d26" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_152[label="@HUB_pyramidbox_lite_mobile_mask@glass_block4_3.w_0"]
   node_139[label="conv2d9" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_120[label="relu4" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_124[label="@HUB_pyramidbox_lite_mobile_mask@relu_2.tmp_0"]
   node_144[label="elementwise_add10" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_180[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn5_2.b_0"]
   node_116[label="scale2" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_156[label="conv2d13" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_187[label="@HUB_pyramidbox_lite_mobile_mask@relu_7.tmp_0"]
   node_194[label="@HUB_pyramidbox_lite_mobile_mask@relu_3.tmp_0"]
   node_114[label="@HUB_pyramidbox_lite_mobile_mask@crelu_bn.b_0"]
   node_111[label="@HUB_pyramidbox_lite_mobile_mask@image"]
   node_137[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn3_3.b_0"]
   node_112[label="conv2d1" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_175[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn4_1.b_0"]
   node_115[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_0.tmp_2"]
   node_142[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn4_3.b_0"]
   node_160[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2"]
   node_153[label="@HUB_pyramidbox_lite_mobile_mask@relu_12.tmp_0"]
   node_123[label="@HUB_pyramidbox_lite_mobile_mask@age_block1_3.w_0"]
   node_110[label="feed"]
   node_198[label="@HUB_pyramidbox_lite_mobile_mask@relu_5.tmp_0"]
   node_117[label="@HUB_pyramidbox_lite_mobile_mask@scale_0.tmp_0"]
   node_129[label="@HUB_pyramidbox_lite_mobile_mask@relu_4.tmp_0"]
   node_130[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn2_3.b_0"]
   node_174[label="@HUB_pyramidbox_lite_mobile_mask@glass_block4_1.w_0"]
   node_132[label="elementwise_add7" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_125[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn1_3.b_0"]
   node_196[label="@HUB_pyramidbox_lite_mobile_mask@age_block3_1.w_0"]
   node_126[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_3.tmp_2"]
   node_128[label="@HUB_pyramidbox_lite_mobile_mask@age_block2_3.w_0"]
   node_155[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_18.tmp_2"]
   node_150[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_15.tmp_2"]
   node_188[label="conv2d22" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_190[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn4_2.b_0"]
   node_157[label="@HUB_pyramidbox_lite_mobile_mask@glass_block5_3.w_0"]
   node_209[label="conv2d28" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_171[label="@HUB_pyramidbox_lite_mobile_mask@glass_block5_1.w_0"]
   node_191[label="conv2d23" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_158[label="@HUB_pyramidbox_lite_mobile_mask@relu_14.tmp_0"]
   node_182[label="@HUB_pyramidbox_lite_mobile_mask@glass_block4_2.w_0"]
   node_162[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1"]
   node_163[label="@HUB_pyramidbox_lite_mobile_mask@save_infer_model/scale_0"]
   node_164[label="fetch15" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_165[label="fetch"]
   node_154[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn4_3.b_0"]
   node_161[label="softmax14" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_166[label="conv2d16" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_167[label="@HUB_pyramidbox_lite_mobile_mask@relu_13.tmp_0"]
   node_169[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn5_2.b_0"]
   node_121[label="@HUB_pyramidbox_lite_mobile_mask@relu_0.tmp_0"]
   node_172[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn5_1.b_0"]
   node_173[label="conv2d18" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_177[label="conv2d19" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_189[label="@HUB_pyramidbox_lite_mobile_mask@age_block4_2.w_0"]
   node_131[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_6.tmp_2"]
   node_179[label="@HUB_pyramidbox_lite_mobile_mask@age_block5_2.w_0"]
   node_159[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn5_3.b_0"]
   node_176[label="@HUB_pyramidbox_lite_mobile_mask@relu_11.tmp_0"]
   node_181[label="conv2d20" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_146[label="conv2d11" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_183[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn4_2.b_0"]
   node_186[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn4_1.b_0"]
   node_214[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn3_2.b_0"]
   node_178[label="@HUB_pyramidbox_lite_mobile_mask@relu_9.tmp_0"]
   node_184[label="conv2d21" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_185[label="@HUB_pyramidbox_lite_mobile_mask@age_block4_1.w_0"]
   node_109[label="feed0" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_192[label="@HUB_pyramidbox_lite_mobile_mask@age_block2_1.w_0"]
   node_195[label="conv2d24" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_197[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn3_1.b_0"]
   node_199[label="conv2d25" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_200[label="@HUB_pyramidbox_lite_mobile_mask@relu_1.tmp_0"]
   node_202[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn1_2.b_0"]
   node_204[label="@HUB_pyramidbox_lite_mobile_mask@age_block2_2.w_0"]
   node_110->node_109
   node_109->node_111
   node_113->node_112
   node_111->node_112
   node_114->node_112
   node_112->node_115
   node_115->node_116
   node_116->node_117
   node_115->node_118
   node_117->node_118
   node_118->node_119
   node_119->node_120
   node_120->node_121
   node_123->node_122
   node_124->node_122
   node_125->node_122
   node_122->node_126
   node_128->node_127
   node_129->node_127
   node_130->node_127
   node_127->node_131
   node_131->node_132
   node_126->node_132
   node_132->node_133
   node_135->node_134
   node_136->node_134
   node_137->node_134
   node_134->node_138
   node_140->node_139
   node_141->node_139
   node_142->node_139
   node_139->node_143
   node_143->node_144
   node_138->node_144
   node_144->node_145
   node_147->node_146
   node_148->node_146
   node_149->node_146
   node_146->node_150
   node_152->node_151
   node_153->node_151
   node_154->node_151
   node_151->node_155
   node_157->node_156
   node_158->node_156
   node_159->node_156
   node_156->node_160
   node_162->node_161
   node_161->node_163
   node_163->node_164
   node_164->node_165
   node_167->node_166
   node_168->node_166
   node_169->node_166
   node_166->node_158
   node_155->node_170
   node_171->node_170
   node_172->node_170
   node_170->node_167
   node_150->node_173
   node_174->node_173
   node_175->node_173
   node_173->node_176
   node_178->node_177
   node_179->node_177
   node_180->node_177
   node_177->node_148
   node_176->node_181
   node_182->node_181
   node_183->node_181
   node_181->node_153
   node_138->node_184
   node_185->node_184
   node_186->node_184
   node_184->node_187
   node_187->node_188
   node_189->node_188
   node_190->node_188
   node_188->node_141
   node_126->node_191
   node_192->node_191
   node_193->node_191
   node_191->node_194
   node_133->node_195
   node_196->node_195
   node_197->node_195
   node_195->node_198
   node_200->node_199
   node_201->node_199
   node_202->node_199
   node_199->node_124
   node_194->node_203
   node_204->node_203
   node_205->node_203
   node_203->node_129
   node_121->node_206
   node_207->node_206
   node_208->node_206
   node_206->node_200
   node_145->node_209
   node_210->node_209
   node_211->node_209
   node_209->node_178
   node_198->node_212
   node_213->node_212
   node_214->node_212
   node_212->node_136
   node_216->node_215
   node_160->node_215
   node_217->node_215
   node_215->node_162
} // end G
[I  3/12 13:11: 2.485 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: type_layout_cast_pass
[I  3/12 13:11: 2.485 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: argument_type_display_pass
[I  3/12 13:11: 2.485 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: argument_type_display_pass
[I  3/12 13:11: 2.485 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: variable_place_inference_pass
[I  3/12 13:11: 2.487 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: variable_place_inference_pass
[I  3/12 13:11: 2.487 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: argument_type_display_pass
[I  3/12 13:11: 2.487 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: argument_type_display_pass
[I  3/12 13:11: 2.487 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: runtime_context_assign_pass
[I  3/12 13:11: 2.487 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: runtime_context_assign_pass
[I  3/12 13:11: 2.487 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: argument_type_display_pass
[I  3/12 13:11: 2.487 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: argument_type_display_pass
[I  3/12 13:11: 2.487 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: memory_optimize_pass
[I  3/12 13:11: 2.487 ...te/lite/core/mir/memory_optimize_pass.cc:104 CollectLifeCycleByDevice] There are 1 types device var.
[I  3/12 13:11: 2.487 ...te/lite/core/mir/memory_optimize_pass.cc:153 MakeReusePlan] cluster: @HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1
[I  3/12 13:11: 2.487 ...te/lite/core/mir/memory_optimize_pass.cc:153 MakeReusePlan] cluster: @HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2
[I  3/12 13:11: 2.487 ...te/lite/core/mir/memory_optimize_pass.cc:153 MakeReusePlan] cluster: @HUB_pyramidbox_lite_mobile_mask@tmp_0
[I  3/12 13:11: 2.487 ...te/lite/core/mir/memory_optimize_pass.cc:153 MakeReusePlan] cluster: @HUB_pyramidbox_lite_mobile_mask@batch_norm_3.tmp_2
[I  3/12 13:11: 2.491 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: memory_optimize_pass
[I  3/12 13:11: 2.491 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: npu_subgraph_pass
[I  3/12 13:11: 2.491 ...ngming/Paddle-Lite/lite/core/optimizer.h:177 RunPasses]    - Skip npu_subgraph_pass because the target or kernel does not match.
[I  3/12 13:11: 2.491 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: xpu_subgraph_pass
[I  3/12 13:11: 2.491 ...ngming/Paddle-Lite/lite/core/optimizer.h:177 RunPasses]    - Skip xpu_subgraph_pass because the target or kernel does not match.
dot:
digraph G {
   node_326[label="@HUB_pyramidbox_lite_mobile_mask@fc1_glass.b_0"]
   node_325[label="@HUB_pyramidbox_lite_mobile_mask@fc1_glass.w_0"]
   node_324[label="fc30" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_322[label="@HUB_pyramidbox_lite_mobile_mask@age_block3_2.w_0"]
   node_320[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn5_1.b_0"]
   node_319[label="@HUB_pyramidbox_lite_mobile_mask@age_block5_1.w_0"]
   node_317[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn1_1.b_0"]
   node_321[label="conv2d29" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_316[label="@HUB_pyramidbox_lite_mobile_mask@age_block1_1.w_0"]
   node_315[label="conv2d27" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_314[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn2_2.b_0"]
   node_231[label="conv2d5" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_256[label="@HUB_pyramidbox_lite_mobile_mask@age_block5_3.w_0"]
   node_252[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1(11)"]
   node_249[label="@HUB_pyramidbox_lite_mobile_mask@age_block4_3.w_0"]
   node_243[label="conv2d8" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_277[label="@HUB_pyramidbox_lite_mobile_mask@glass_block5_2.w_0"]
   node_260[label="conv2d12" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_242[label="@HUB_pyramidbox_lite_mobile_mask@tmp_0"]
   node_258[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn5_3.b_0"]
   node_259[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1(15)"]
   node_310[label="@HUB_pyramidbox_lite_mobile_mask@age_block1_2.w_0"]
   node_240[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2(5)"]
   node_276[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2(19)"]
   node_227[label="concat3" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_222[label="@HUB_pyramidbox_lite_mobile_mask@crelu.w_0"]
   node_236[label="conv2d6" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_302[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn2_1.b_0"]
   node_312[label="conv2d26" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_261[label="@HUB_pyramidbox_lite_mobile_mask@glass_block4_3.w_0"]
   node_248[label="conv2d9" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_229[label="relu4" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_309[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2(1)"]
   node_225[label="scale2" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_265[label="conv2d13" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_223[label="@HUB_pyramidbox_lite_mobile_mask@crelu_bn.b_0"]
   node_220[label="@HUB_pyramidbox_lite_mobile_mask@image"]
   node_254[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2(12)"]
   node_230[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1(0)"]
   node_266[label="@HUB_pyramidbox_lite_mobile_mask@glass_block5_3.w_0"]
   node_300[label="conv2d23" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_246[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn3_3.b_0"]
   node_221[label="conv2d1" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_284[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn4_1.b_0"]
   node_244[label="@HUB_pyramidbox_lite_mobile_mask@age_block3_3.w_0"]
   node_250[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2(10)"]
   node_224[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_0.tmp_2"]
   node_251[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn4_3.b_0"]
   node_269[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2"]
   node_253[label="elementwise_add10" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_289[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn5_2.b_0"]
   node_232[label="@HUB_pyramidbox_lite_mobile_mask@age_block1_3.w_0"]
   node_247[label="@HUB_pyramidbox_lite_mobile_mask@tmp_0(8)"]
   node_291[label="@HUB_pyramidbox_lite_mobile_mask@glass_block4_2.w_0"]
   node_271[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1"]
   node_226[label="@HUB_pyramidbox_lite_mobile_mask@scale_0.tmp_0"]
   node_219[label="feed"]
   node_287[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1(13)"]
   node_262[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1(17)"]
   node_267[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1(20)"]
   node_235[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_3.tmp_2"]
   node_245[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2(7)"]
   node_305[label="@HUB_pyramidbox_lite_mobile_mask@age_block3_1.w_0"]
   node_237[label="@HUB_pyramidbox_lite_mobile_mask@age_block2_3.w_0"]
   node_239[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn2_3.b_0"]
   node_283[label="@HUB_pyramidbox_lite_mobile_mask@glass_block4_1.w_0"]
   node_238[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1(4)"]
   node_307[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1(6)"]
   node_272[label="@HUB_pyramidbox_lite_mobile_mask@save_infer_model/scale_0"]
   node_233[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1(2)"]
   node_273[label="fetch15" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_274[label="fetch"]
   node_263[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn4_3.b_0"]
   node_270[label="softmax14" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_275[label="conv2d16" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_278[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn5_2.b_0"]
   node_279[label="conv2d17" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_281[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn5_1.b_0"]
   node_257[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2(14)"]
   node_282[label="conv2d18" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_241[label="elementwise_add7" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_234[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn1_3.b_0"]
   node_285[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2(16)"]
   node_306[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn3_1.b_0"]
   node_286[label="conv2d19" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_298[label="@HUB_pyramidbox_lite_mobile_mask@age_block4_2.w_0"]
   node_288[label="@HUB_pyramidbox_lite_mobile_mask@age_block5_2.w_0"]
   node_268[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn5_3.b_0"]
   node_290[label="conv2d20" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_255[label="conv2d11" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_292[label="@HUB_pyramidbox_lite_mobile_mask@glass_block_bn4_2.b_0"]
   node_295[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn4_1.b_0"]
   node_323[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn3_2.b_0"]
   node_293[label="conv2d21" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_294[label="@HUB_pyramidbox_lite_mobile_mask@age_block4_1.w_0"]
   node_318[label="conv2d28" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_280[label="@HUB_pyramidbox_lite_mobile_mask@glass_block5_1.w_0"]
   node_296[label="@HUB_pyramidbox_lite_mobile_mask@fc_0.tmp_1(9)"]
   node_297[label="conv2d22" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_264[label="@HUB_pyramidbox_lite_mobile_mask@tmp_0(18)"]
   node_299[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn4_2.b_0"]
   node_218[label="feed0" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_301[label="@HUB_pyramidbox_lite_mobile_mask@age_block2_1.w_0"]
   node_228[label="@HUB_pyramidbox_lite_mobile_mask@concat_0.tmp_0"]
   node_303[label="@HUB_pyramidbox_lite_mobile_mask@batch_norm_21.tmp_2(3)"]
   node_304[label="conv2d24" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_308[label="conv2d25" shape="box" style="filled" color="black" fillcolor="yellow"]
   node_311[label="@HUB_pyramidbox_lite_mobile_mask@age_block_bn1_2.b_0"]
   node_313[label="@HUB_pyramidbox_lite_mobile_mask@age_block2_2.w_0"]
   node_219->node_218
   node_218->node_220
   node_222->node_221
   node_220->node_221
   node_223->node_221
   node_221->node_224
   node_224->node_225
   node_225->node_226
   node_224->node_227
   node_226->node_227
   node_227->node_228
   node_228->node_229
   node_229->node_230
   node_232->node_231
   node_233->node_231
   node_234->node_231
   node_231->node_235
   node_237->node_236
   node_238->node_236
   node_239->node_236
   node_236->node_240
   node_240->node_241
   node_235->node_241
   node_241->node_242
   node_244->node_243
   node_245->node_243
   node_246->node_243
   node_243->node_247
   node_249->node_248
   node_250->node_248
   node_251->node_248
   node_248->node_252
   node_252->node_253
   node_247->node_253
   node_253->node_254
   node_256->node_255
   node_257->node_255
   node_258->node_255
   node_255->node_259
   node_261->node_260
   node_262->node_260
   node_263->node_260
   node_260->node_264
   node_266->node_265
   node_267->node_265
   node_268->node_265
   node_265->node_269
   node_271->node_270
   node_270->node_272
   node_272->node_273
   node_273->node_274
   node_276->node_275
   node_277->node_275
   node_278->node_275
   node_275->node_267
   node_264->node_279
   node_280->node_279
   node_281->node_279
   node_279->node_276
   node_259->node_282
   node_283->node_282
   node_284->node_282
   node_282->node_285
   node_287->node_286
   node_288->node_286
   node_289->node_286
   node_286->node_257
   node_285->node_290
   node_291->node_290
   node_292->node_290
   node_290->node_262
   node_247->node_293
   node_294->node_293
   node_295->node_293
   node_293->node_296
   node_296->node_297
   node_298->node_297
   node_299->node_297
   node_297->node_250
   node_235->node_300
   node_301->node_300
   node_302->node_300
   node_300->node_303
   node_242->node_304
   node_305->node_304
   node_306->node_304
   node_304->node_307
   node_309->node_308
   node_310->node_308
   node_311->node_308
   node_308->node_233
   node_303->node_312
   node_313->node_312
   node_314->node_312
   node_312->node_238
   node_230->node_315
   node_316->node_315
   node_317->node_315
   node_315->node_309
   node_254->node_318
   node_319->node_318
   node_320->node_318
   node_318->node_287
   node_307->node_321
   node_322->node_321
   node_323->node_321
   node_321->node_245
   node_325->node_324
   node_269->node_324
   node_326->node_324
   node_324->node_271
} // end G
[I  3/12 13:11: 2.492 ...te/lite/core/mir/generate_program_pass.h:37 GenProgram] insts.size 31
[I  3/12 13:11: 2.508 ...e-Lite/lite/model_parser/model_parser.cc:589 SaveModelNaive] Save naive buffer model in 'model' successfully
model  __model__  model.nb  opt  __params__

而後咱們在test_program/mask_detector/路徑下能夠看到model.nb ，這個是已經使用opt工具轉化好的模型，下載下來後放到樹莓派裏進行模型替換，這裏就完成了口罩模型的替換。

轉化完mask模型還須要轉化face模型，方式是與轉化mask模型是同樣的。這個項目是對人臉和口罩分別都有對應的模型的，目前2.3.0版本的opt工具在轉化face模型時會報錯，這個我已經提交了issue，等待修復。修復好後更新模型的方式我已經寫入FAQ中～

In[13]

%cd /home/aistudio/test_program/pyramidbox_lite
!ls && pwd

#使用opt工具進行模型轉化 將__model__ 和 __params__ 轉化爲model.nb (這裏會報錯 等待官方修復opt工具)
!./opt --model_file=./__model__ --param_file=./__params__ --optimize_out_type=naive_buffer --optimize_out=model

/home/aistudio/test_program/pyramidbox_lite
__model__  opt	__params__
/home/aistudio/test_program/pyramidbox_lite
[W  3/12 13:12:36.990 ...dle/hongming/Paddle-Lite/lite/api/opt.cc:138 RunOptimize] Load combined-param model. Option model_dir will be ignored
[I  3/12 13:12:36.990 ...hongming/Paddle-Lite/lite/api/cxx_api.cc:244 Build] Load model from file.
[I  3/12 13:12:37. 10 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_quant_dequant_fuse_pass
[I  3/12 13:12:37. 16 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_quant_dequant_fuse_pass
[I  3/12 13:12:37. 16 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: weight_quantization_preprocess_pass
[I  3/12 13:12:37. 17 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: weight_quantization_preprocess_pass
[I  3/12 13:12:37. 17 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_conv_elementwise_fuse_pass
[I  3/12 13:12:37. 20 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 12 subgraph
[I  3/12 13:12:37. 21 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_conv_elementwise_fuse_pass
[I  3/12 13:12:37. 21 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_conv_bn_fuse_pass
[I  3/12 13:12:37. 32 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 25 subgraph
[I  3/12 13:12:37. 35 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_conv_bn_fuse_pass
[I  3/12 13:12:37. 35 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_conv_elementwise_fuse_pass
[I  3/12 13:12:37. 39 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_conv_elementwise_fuse_pass
[I  3/12 13:12:37. 39 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_conv_activation_fuse_pass
[I  3/12 13:12:37. 43 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 28 subgraph
[I  3/12 13:12:37. 52 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_conv_activation_fuse_pass
[I  3/12 13:12:37. 52 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_var_conv_2d_activation_fuse_pass
[I  3/12 13:12:37. 52 ...ngming/Paddle-Lite/lite/core/optimizer.h:177 RunPasses]    - Skip lite_var_conv_2d_activation_fuse_pass because the target or kernel does not match.
[I  3/12 13:12:37. 52 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_fc_fuse_pass
[I  3/12 13:12:37. 52 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_fc_fuse_pass
[I  3/12 13:12:37. 52 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_shuffle_channel_fuse_pass
[I  3/12 13:12:37. 52 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_shuffle_channel_fuse_pass
[I  3/12 13:12:37. 52 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_transpose_softmax_transpose_fuse_pass
[I  3/12 13:12:37. 53 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_transpose_softmax_transpose_fuse_pass
[I  3/12 13:12:37. 53 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_interpolate_fuse_pass
[I  3/12 13:12:37. 53 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: lite_interpolate_fuse_pass
[I  3/12 13:12:37. 53 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: identity_scale_eliminate_pass
[I  3/12 13:12:37. 53 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 3 subgraph
[I  3/12 13:12:37. 54 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: identity_scale_eliminate_pass
[I  3/12 13:12:37. 54 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: elementwise_mul_constant_eliminate_pass
[I  3/12 13:12:37. 54 ...ngming/Paddle-Lite/lite/core/optimizer.h:181 RunPasses] == Finished running: elementwise_mul_constant_eliminate_pass
[I  3/12 13:12:37. 54 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: lite_sequence_pool_concat_fuse_pass
[I  3/12 13:12:37. 54 ...ngming/Paddle-Lite/lite/core/optimizer.h:177 RunPasses]    - Skip lite_sequence_pool_concat_fuse_pass because the target or kernel does not match.
[I  3/12 13:12:37. 54 ...ngming/Paddle-Lite/lite/core/optimizer.h:164 RunPasses] == Running pass: static_kernel_pick_pass
[F  3/12 13:12:37. 57 ...hongming/Paddle-Lite/lite/core/kernel.cc:44 GetOutputDeclType] Check failed: type: no type registered for kernel [multiclass_nms/def] output argument [Index]
Aborted (core dumped)

5 FAQ

爲何項目中的模型是__model__.nb和params.nb兩個文件呢？

由於項目中使用的模型是Paddle-Lite v2.2.0版本以前的模型，v2.3.0之後將兩個文件合併成一個文件，方便使用。等待v3.0.0後加載老版本模型的APi也將移除。

爲何不在項目裏使用最新的模型呢?

由於我在使用opt轉化pyramidbox_lite模型是會報錯,是opt工具的問題,現已經提交了issue等待攻城獅們修復~

後續等待模型可用時應該如何更新模型呢?

由於v2.2.0與v2.3.0的模型形式以及加載模型的API都發生了變化，這裏都須要修改一下。

因此具體步驟以下:

當使用新版本opt工具生成模型後,移動置對應的模型文件夾下
修改run.sh文件,將第18行和21行的../models/face_detection ../models/mask_classification改成../models/face_detection/model.nb ../models/mask_classification/model.nb
修改源代碼,使用新版本加載模型的api,打開mask_detection.cc源碼,搜索config.set_model_dir,有3處分別在115,162,253,300行,將其全都修改成set_model_from_file 附贈上2.3.0的API

而後的使用方式和以前同樣，目前的效果還闊以接受。使用體驗上畫面會有些延遲，也是由於樹莓派的CPU對視頻流的處理能力仍是有限的。

若是我是armv7hf/32位系統的該怎麼運行項目？

對於armv7hf/32位系統須要對應的Paddle-Lite庫，項目裏是armv8針對64位的系統的。armv7hf/32位系統須要自行編譯

編譯方式以下：

sudo apt update
#安裝源碼編譯Paddle-lite時必要的工具
sudo apt-get install gcc g++ make wget python unzip

CMake3.10前面配置軟件環境時已經安裝好了的就不用再次安裝了，沒有安裝的參照上面1.3運行環境進行安裝

#下載Paddle-Lite源碼（兩種方式）
方式一：
git clone https://github.com/PaddlePaddle/Paddle-Lite.git
cd Paddle-Lite
#檢查分支 默認是develop分支
git checkout <release-version-tag>
——————————————————————————————————————————————————————————————————————————
（推薦）方式二：
這裏下載2.3.0 release版本
wget https://github.com/PaddlePaddle/Paddle-Lite/archive/v2.3.0.zip

而後解壓後cd Paddle-Lite

最後開始編譯：
./lite/tools/build.sh \
  --build_extra=OFF \
  --arm_os=armlinux \
  --arm_abi=armv7hf \
  --arm_lang=gcc \
  tiny_publish

關於具體參數可參考官方文檔：連接

編譯好後最後生成的文件位於build.lite.armlinux.armv7hf.gcc。取出include和lib放在樹莓派進行對應的文件替換便可。別忘了還須要在run.sh裏將第四行TARGET_ARCH_ABI=armv8註釋掉，取消第5行#TARGET_ARCH_ABI=armv7hf的註釋

最後以爲有幫助的話別忘了fork一下喔

能再收藏關注三連就更好了啊哈哈哈～

使用AI Studio一鍵上手實踐項目吧：https://aistudio.baidu.com/aistudio/projectdetail/315730

下載安裝命令

## CPU版本安裝命令
pip install -f https://paddlepaddle.org.cn/pip/oschina/cpu paddlepaddle

## GPU版本安裝命令
pip install -f https://paddlepaddle.org.cn/pip/oschina/gpu paddlepaddle-gpu

>> 訪問 PaddlePaddle 官網，瞭解更多相關內容。

相關標籤/搜索

每日一句

每一个你不满意的现在，都有一个你没有努力的曾经。