tensorflow目标检测边训练边测试（SSD）

最新推荐文章于 2024-07-03 11:42:50 发布
最新推荐文章于 2024-07-03 11:42:50 发布
阅读量1.1k 收藏 2
点赞数
文章标签： tensorflow mobilenet v2 深度学习
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/qq_35975447/article/details/105950332
版权

为什么写这篇呢，因为近期使用tensorflow的一些发现吧。好记性不如烂笔头！希望能为以后解惑。

目录

0.环境

1.文件准备

2.train

3.eval

4.tensorboard显示

0.环境

ubuntu
python3.6
tensorflow/models

1.文件准备

参考tensorflow lite：prepare our data, train on it and convert the model from ckpt pb to tflite format，前4步中的准备。

2.train

训练ssd mobilenet v2的模型，训练可以使用以下命令：

python3 train.py --logtostderr --train_dir=./logs --pipeline_config_path=./ssd_mobilenet_v2.config

3.eval

测试可以使用以下命令：

python3 eval.py --logtostderr --eval_dir=logs/test_logs/ --pipeline_config_path=ssd_mobilenet_v2.config --checkpoint_dir=logs/

 

INFO:tensorflow:Starting evaluation at 2020-05-06-06:15:46
INFO:tensorflow:Starting evaluation at 2020-05-06-06:15:46
INFO:tensorflow:Restoring parameters from /logs/model.ckpt-2023
INFO:tensorflow:Restoring parameters from /logs/model.ckpt-2023
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-0.
INFO:tensorflow:Detection visualizations written to summary with tag image-0.
WARNING:root:image b'1587371130748.jpg' does not have groundtruth difficult flag specified
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-1.
INFO:tensorflow:Detection visualizations written to summary with tag image-1.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-2.
INFO:tensorflow:Detection visualizations written to summary with tag image-2.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-3.
INFO:tensorflow:Detection visualizations written to summary with tag image-3.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-4.
INFO:tensorflow:Detection visualizations written to summary with tag image-4.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-5.
INFO:tensorflow:Detection visualizations written to summary with tag image-5.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-6.
INFO:tensorflow:Detection visualizations written to summary with tag image-6.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-7.
INFO:tensorflow:Detection visualizations written to summary with tag image-7.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-8.
INFO:tensorflow:Detection visualizations written to summary with tag image-8.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-9.
INFO:tensorflow:Detection visualizations written to summary with tag image-9.
INFO:tensorflow:Running eval ops batch 100/124
INFO:tensorflow:Running eval ops batch 100/124
INFO:tensorflow:Running eval batches done.
INFO:tensorflow:Running eval batches done.
INFO:tensorflow:# success: 124
INFO:tensorflow:# success: 124
INFO:tensorflow:# skipped: 0
INFO:tensorflow:# skipped: 0
INFO:tensorflow:Writing metrics to tf summary.
INFO:tensorflow:Writing metrics to tf summary.
INFO:tensorflow:Losses/Loss/classification_loss: 6.880040
INFO:tensorflow:Losses/Loss/classification_loss: 6.880040
INFO:tensorflow:Losses/Loss/localization_loss: 1.355176
INFO:tensorflow:Losses/Loss/localization_loss: 1.355176
INFO:tensorflow:PascalBoxes_PerformanceByCategory/AP@0.5IOU/b'face': 0.887477
INFO:tensorflow:PascalBoxes_PerformanceByCategory/AP@0.5IOU/b'face': 0.887477
INFO:tensorflow:PascalBoxes_Precision/mAP@0.5IOU: 0.887477
INFO:tensorflow:PascalBoxes_Precision/mAP@0.5IOU: 0.887477
INFO:tensorflow:Metrics written to tf summary.
INFO:tensorflow:Metrics written to tf summary.
INFO:tensorflow:Starting evaluation at 2020-05-06-06:20:46
INFO:tensorflow:Starting evaluation at 2020-05-06-06:20:46
INFO:tensorflow:Found already evaluated checkpoint. Will try again in 300 seconds
INFO:tensorflow:Found already evaluated checkpoint. Will try again in 300 seconds
INFO:tensorflow:Starting evaluation at 2020-05-06-06:25:46
INFO:tensorflow:Starting evaluation at 2020-05-06-06:25:46
INFO:tensorflow:Restoring parameters from /logs/model.ckpt-2201
INFO:tensorflow:Restoring parameters from /logs/model.ckpt-2201
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-0.
INFO:tensorflow:Detection visualizations written to summary with tag image-0.
WARNING:root:image b'1587371130748.jpg' does not have groundtruth difficult flag specified
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-1.
INFO:tensorflow:Detection visualizations written to summary with tag image-1.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-2.
INFO:tensorflow:Detection visualizations written to summary with tag image-2.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-3.
INFO:tensorflow:Detection visualizations written to summary with tag image-3.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-4.
INFO:tensorflow:Detection visualizations written to summary with tag image-4.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-5.
INFO:tensorflow:Detection visualizations written to summary with tag image-5.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-6.
INFO:tensorflow:Detection visualizations written to summary with tag image-6.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-7.
INFO:tensorflow:Detection visualizations written to summary with tag image-7.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-8.
INFO:tensorflow:Detection visualizations written to summary with tag image-8.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Creating detection visualizations.
INFO:tensorflow:Detection visualizations written to summary with tag image-9.
INFO:tensorflow:Detection visualizations written to summary with tag image-9.
INFO:tensorflow:Running eval ops batch 100/124
INFO:tensorflow:Running eval ops batch 100/124
INFO:tensorflow:Running eval batches done.
INFO:tensorflow:Running eval batches done.
INFO:tensorflow:# success: 124
INFO:tensorflow:# success: 124
INFO:tensorflow:# skipped: 0
INFO:tensorflow:# skipped: 0
INFO:tensorflow:Writing metrics to tf summary.
INFO:tensorflow:Writing metrics to tf summary.
INFO:tensorflow:Losses/Loss/classification_loss: 5.544095
INFO:tensorflow:Losses/Loss/classification_loss: 5.544095
INFO:tensorflow:Losses/Loss/localization_loss: 1.116960
INFO:tensorflow:Losses/Loss/localization_loss: 1.116960
INFO:tensorflow:PascalBoxes_PerformanceByCategory/AP@0.5IOU/b'face': 0.902882
INFO:tensorflow:PascalBoxes_PerformanceByCategory/AP@0.5IOU/b'face': 0.902882
INFO:tensorflow:PascalBoxes_Precision/mAP@0.5IOU: 0.902882
INFO:tensorflow:PascalBoxes_Precision/mAP@0.5IOU: 0.902882
INFO:tensorflow:Metrics written to tf summary.
INFO:tensorflow:Metrics written to tf summary.
INFO:tensorflow:Starting evaluation at 2020-05-06-06:30:46
INFO:tensorflow:Starting evaluation at 2020-05-06-06:30:46
INFO:tensorflow:Found already evaluated checkpoint. Will try again in 300 seconds
INFO:tensorflow:Found already evaluated checkpoint. Will try again in 300 seconds
INFO:tensorflow:Finished evaluation!
INFO:tensorflow:Finished evaluation!

如果我们不想边训练边测试的话，我们可以使用如下命令（将我们想测试的单个epoch的3个ckpt复制与checkpoint复制到test_logs目录下，然后修改checkpoint当前ckpt为指定epoch）：

python3 eval.py --logtostderr --eval_dir=logs/test_logs/ --pipeline_config_path=ssd_mobilenet_v2.config --checkpoint_dir=logs/test_logs/

 在手动测试单个ckpt时，我们可以在tensorflow等待的300s过程（ Will try again in 300 seconds）中添加新的ckpt过去，并对应修改checkpoint就可以了。

如果完成测试后，我们重新对新生成的ckpt继续测试，tensorboard可以连续地显示测试曲线。

4.tensorboard显示

cd进入logs目录的前一级目录下，运行下面的命令，我们就可以同时看到train与eval的曲线了。

tensorboard --logdir ./

 

 

wait a minutes
关注
Tensorflow 一边训练一边验证（training and validation）
董煎饼的沙滩
07-20 1万+
在使用tensorflow来训练一个模型的时候，有时候需要依靠验证集来判断模型是否已经过拟合，是否需要停止训练。 1.首先想到的是用tf.placeholder()载入不同的数据来进行计算，比如 def inference(input_): """ this is where you put your graph. the following is just an e...
Tensorflow中怎么使用queue读取数据的情况下，在同一个session中边训练边测试
silence1214的专栏
09-07 3159
通常情况下，我们在训练的时候都想边训练边找来一批验证数据看看在验证数据上的性能。但是由于tensorflow中如果使用queue的方法来读取数据的话，这方法就不好实现了，主要问题是由于queue读取数据的时候，我们的graph的构建就不是使用placeholder的方法而是，那么在同一个session中就没办法换输入数据了。为了更好的描述这个问题，我先给出代码：#使用queue方法得到的数据 tra
caffe 的边训练边测试
activeforcus的专栏
12-17 1322
个人认为需要改以下两个地方：训练文件中增加测试层                                                         在solver.prototxt中设定迭代多少次进行测试 train_val.prototxt ###### test layer, ignore_label layer {   name: "accuracy"   type
基于tensorflow2的目标检测完整实现过程
最新发布
qq_45055172的博客
07-03 1208
基于tensorflow框架，总结包含图片标注、公开数据集下载、数据格式转换、目标检测API安装、预训练模型下载、模型训练和调用。目标检测快速落地全流程总结。
tensorflow目标检测任务训练时一直测试
聿默的博客
06-29 259
写这篇的原因，是测试几个中间epoch后就会停止测试。还需要手动运行测试命令。 由于tensorflow目标检测运行eval时，会不断等待300s： 然后测试新的模型。如何没有生成新的模型多次或者是其他的什么机制，使得测试停止，那就会输出完成测试： 那怎么才能训练的时候一直测试呢，写个shell-while，不用修改源代码，不香吗？ 1.编写eval.sh eval.sh #!/bin/bash while(1) do CUDA_VISIBLE_DEVICES=3 python
同时完成“训练和测试”(3000fps)
那年聪聪
03-30 1100
同时完成“训练和测试” 图像集训练 图像数据的测试
tensorflow目标检测代码
12-01
在这个"tensorflow目标检测代码"项目中，我们看到它支持Python 2.7版本，这意味着我们需要使用旧版的TensorFlow库，因为TensorFlow 2.x默认不支持Python 2。在设置环境中，确保安装了正确的TensorFlow版本，以及相关...
基于SSD-Tensorflow目标检测+医学影像的消化内镜活检钳源码+部署教程文档+全部数据+训练好的模型（高分项目）.zip
04-20
基于SSD-Tensorflow目标检测+医学影像的消化内镜活检钳源码+部署教程文档+全部数据+训练好的模型（高分项目）.zip基于SSD-Tensorflow目标检测+医学影像的消化内镜活检钳源码+部署教程文档+全部数据+训练好的模型...
在tensorflow下实现SSD的目标检测
MALLAIN的博客
11-19 661
训练自己的数据集环境配置1.生成数据集2.训练数据3.根据自己生成的模型进行预测4.总结 环境配置 tensorflow-gpu == 1.15.0 keras == 2.3.1 (可以不用安装) CUDN == 10.0 CUDNN == 7.6 1.生成数据集 在SSD目标检测中需要将VOC数据集转化成.tf格式。 运行tf_convert_data.py文件 #需要编辑的代码 import tensorflow as tf from datasets import pascalvoc_to_tfr
tensorflow2.0训练目标检测模型
weixin_48672949的博客
07-21 2981
1.环境搭建与软件安装 操作系统：win10 64位 内存 ：8G Anaconda3-5.0.1 (自带python3.6.5) tensorflow2.5.0 1.1 Anaconda3-5.0.1版本安装包下载、安装流程、环境配置 参考https://www.freesion.com/article/93321279940/ 验证：点开菜单栏—>Anaconda3—>Anaconda prompt—>以管理员身份运行—>输入python，如下图所示，Anacon.
SSD-Tensorflow 目标检测
weixin_38332967的博客
07-31 2697
参考网址：https://blog.csdn.net/weixin_39881922/article/details/80569803                   https://blog.csdn.net/liuyan20062010/article/details/78905517                   https://blog.csdn.net/ei1990/art...
pytorch_SSD训练评估自己的数据集
Lighter_Go的博客
06-23 1106
目录1. 模型环境2. 数据集处理2.1 数据格式转换yolo_to_voc2.2 数据划分2.3 数据集结构3. 模型代码配置3.1 代码配置3.2 设置自己的代码4. BUG问题 1. 模型环境 利用**ssd.pytorch**模型 运行环境 显卡P5000 ubuntu16.04+CUDA9.2 pytorch1.0.1 , torchvision 0.2.2 2. 数据集处理 2.1 数据格式转换yolo_to_voc 由于开始制作的数据集是为了在yolov3模型下进行训练的yolo格
备注：tensorflow模型训练时同时进行测试，将结果可视化。
tiandd12的博客
09-23 5495
1、首先创建两个placeholder，用来存放测试数据集，和标签。 x = tf.placeholder(tf.float32, shape = [BATCH_SIZE,256,256,3]) y_ = tf.placeholder(tf.int16,shape = [BATCH_SIZE]) 2、创建一个FileWriter ,用于写入测试数据信息 test_writer = tf.
【推荐收藏】【机器学习实战】训练模型（挑战全网最全）
热门推荐
惟忆
01-16 4万+
参照《机器学习实战》第二版 本章探讨的大部分主题对于理解、构建和训练神经网络是至关重要的。 目的在于了解系统是如何工作的，它有助于快速定位到适合的模型、正确的训练算法，以及一套合适的参数。不仅如此，后期还能让你更高效的执行错误调试和错误分析。 我们将从最简单的模型之一 – 线性回归模型开始，介绍两种非常不同的训练模型的方法： 通过“闭式”方程，直接计算出最拟合训练集的模型参数（也就是使训练集上的成本模型最小化的模型参数）。 使用迭代优化的方法，即梯度下降（GD)，逐渐调整模型参数直至训练集上的成本函数.
关于深度学习训练测试过程的通俗理解
weixin_34391445的博客
05-27 2768
深度学习包括训练和测试阶段： 训练：把原料（训练数据），按照丹方（比如yolo网络结构），倒进丹炉（tf 或者caffe框架），使用三味真火（CPU或者GPU），提炼出丹药（模型参数）的过程。 测试：就是把测试数据用模型参数按照上面炼丹的过程走一遍。 转载于:https://www.cnblogs.com/llfctt/p/9097399.html...
Demo-深度学习使用数据集进行训练和测试
u010757019的博客
01-06 2858
一个包含数据输入和预处理流程的使用数据集进行训练和测试的完整例子 import tensorflow as tf train_files = tf.train.match_filenames_once("path/to/train-file-*") test_files = tf.train.match_filenames_once("path/to/test-file-*") # 定义pa...
如何一边训练一遍验证
weixin_44456198的博客
12-05 977
1.首先想到的是用tf.placeholder()载入不同的数据来进行计算，比如 def inference(input_): """ this is where you put your graph. the following is just an example. """ conv1 = tf.layers.conv2d(input_) ...
Tensorflow深度学习入门到实战
06-05
Tensorflow是当今各大互联网首选使用的深度学习框架，最新的Tensorflow2版本提倡的kearas搭建模型的方法，能够快速的实现和部署深度学习模型，应用于自然语言处理、推荐系统等分类和预估技术领域，掌握了Tensorflow技术会成为一个人职场的核心竞争力，本视频系列从实战角度出发，通过一个个的数据建模案例带你掌握这门热门技术
机器学习训练好的模型的使用_一份关于机器学习“模型再训练”的终极指南
weixin_39611049的博客
11-24 1700
机器学习模型的训练，通常是通过学习某一组输入特征与输出目标之间的映射来进行的。一般来说，对于映射的学习是通过优化某些成本函数，来使预测的误差最小化。在训练出最佳模型之后，将其正式发布上线，再根据未来生成的数据生成准确的预测。这些新数据示例可能是用户交互、应用处理或其他软件系统的请求生成的——这取决于模型需要解决的问题。在理想情况下，我们会希望自己的模型在生产环境中进行预测时，能够像使用训练过程中...
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包

打赏作者

wait a minutes

你的鼓励将是我创作的最大动力

¥1 ¥2 ¥4 ¥6 ¥10 ¥20
扫码支付：¥1
获取中
扫码支付

您的余额不足，请更换扫码支付或充值

打赏作者

实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值