About training data, What is bg_img_dir ?
ldfinfontainebleau opened this issue · comments
I am trying to train my hand pose model with your contribution， thanks. When I reading your code, I am a little confused about bg_img_dir(in config.py). Can u tell me what is it, please?
I write my generator and ignore bg_img, it is training now. Everything seems fine.
what dataset can be used in this repo?
what dataset can be used in this repo?
Do you mean data of the training program? I used it on hand key points. My training data just have two parts :
1.hands' pictures * .jpg
2.keypoints *.json
You could write your generator with your own datasets. I think it will be fine. (xiaolian)
what dataset can be used in this repo?
Do you mean data of the training program? I used it on hand key points. My training data just have two parts :
1.hands' pictures * .jpg
2.keypoints *.json
You could write your generator with your own datasets. I think it will be fine. (xiaolian)
thks, but how to get key point efficiently, Is it manually marked?
How do you get the data set? Whether to label yourself or use the official data set? thks
It's private data(about 200000+ imgs, company property). You could search some public data sets, but the public data, I have seen, seems a little rough to me.
I suggest you to look at CMU Panoptic Dataset.
Good Luck!
what dataset can be used in this repo?
Do you mean data of the training program? I used it on hand key points. My training data just have two parts :
1.hands' pictures * .jpg
2.keypoints *.json
You could write your generator with your own datasets. I think it will be fine. (xiaolian)thks, but how to get key point efficiently, Is it manually marked?
Yes, manually and precisely.
How do you get the data set? Whether to label yourself or use the official data set? thks
It's private data(about 200000+ imgs, company property). You could search some public data sets, but the public data, I have seen, seems a little rough to me.
I suggest you to look at CMU Panoptic Dataset.
Good Luck!
Thank you very much for your help, I will try your recommended data set.
@ldfinfontainebleau can you give me step by step how to trainning new data, i alreay have my dataset but don't know how to create format in this project and make it train :(
@ldfinfontainebleau can you give me step by step how to trainning new data, i alreay have my dataset but don't know how to create format in this project and make it train :(
Have you found a data set annotation method? I'm also working on my own data set. Asking for advice~~ thank you ~
step by step... It has been a while. I will give it a try.
- make sure your dataset has two parts : .jpg and landmarks
- I did fix the program to suit my data form
- generate jpg and landmarks into training
this project is just suit its data, I think u should change a little before start training.
I hope I could use English fluently as my Chinese : )
step by step... It has been a while. I will give it a try.
1. make sure your dataset has two parts : .jpg and landmarks 2. I did fix the program to suit my data form 3. generate jpg and landmarks into training this project is just suit its data, I think u should change a little before start training. I hope I could use English fluently as my Chinese : )
thank you~ I make labels.txt contrast what is introduced in Readme like this:
20008.jpg 34.0 153.0 172 501 40.0 242.0 53.0 241.0 102.0 201.0 160.0 158.0 118.0 288.0 170.0 321.0 167.0 240.0 277.0 242.0 371.0 198.0 445.0 171.0 374.0 289.0 444.0 322.0
I train body-pose and data_picture have 12 points in every bbox, but now I'm not sure which files I need to modify in order to train properly.(qidai~~~)
@qqwinn 在config.py里面有heatmap_size热力图大小，越大越准确但参数也越多；cpm_stages同理，是定义cpm阶段数的，一般最少5段，上可以加。这两个参数对于网络我觉得都很重要。需要根据自己的数据来看吧。（手动微笑）
@qqwinn 在config.py里面有heatmap_size热力图大小，越大越准确但参数也越多；cpm_stages同理，是定义cpm阶段数的，一般最少5段，上可以加。这两个参数对于网络我觉得都很重要。需要根据自己的数据来看吧。（手动微笑）
I write my generator and ignore bg_img, it is training now. Everything seems fine.
蟹蟹你~~
I display an "import Ensemble_data_generator error" when I runing run_train.py. Does this file need to be written by myself ? But I don't know exactly how to write this file. Can I refer to your generator code?~
[再次期待 :) :) :) ]
@qqwinn 要看我写的generator？留个联系方式我发给你。: |)
@qqwinn 要看我写的generator？留个联系方式我发给你。: |)
上一个邮箱我貌似填错了，真是不好意思～
qqwinn@aliyun.com 这次没错 ：：|
@qqwinn 要看我写的generator？留个联系方式我发给你。: |)
上一个邮箱我貌似填错了，真是不好意思～
qqwinn@aliyun.com 这次没错 ：：|
没事的，这两天在忙一个项目，还没来得及发给你，晚上发给你。
@qqwinn 要看我写的generator？留个联系方式我发给你。: |)
上一个邮箱我貌似填错了，真是不好意思～
qqwinn@aliyun.com 这次没错 ：：|没事的，这两天在忙一个项目，还没来得及发给你，晚上发给你。
@ldfinfontainebleau 可以给发我一份吗 谢谢 2501026434@qq.com
@ldfinfontainebleau 您好，我在生成数据的时候也遇到了同样的问题，方便的话，可否也将代码共享我一下呀？谢谢啦，邮箱：dx13386842300@gmail.com ,非常感谢！
你好： 这个是别人写的generator.py文件，后续有什么心得可以交流哈～～～ 王琦琦
…
------------------------------------------------------------------ 发件人：mathduanxu notifications@github.com 发送时间：2019年5月7日(星期二) 14:44 收件人：timctho/convolutional-pose-machines-tensorflow convolutional-pose-machines-tensorflow@noreply.github.com 抄 送：qqwinn qqWinn@aliyun.com; Mention mention@noreply.github.com 主 题：Re: [timctho/convolutional-pose-machines-tensorflow] About training data, What is bg_img_dir ? (#48) @ldfinfontainebleau 您好，我在生成数据的时候也遇到了同样的问题，方便的话，可否也将代码共享我一下呀？谢谢啦，邮箱：dx13386842300@gmail.com ,非常感谢！ — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.
我要要一份generator.py,谢谢1903847062@QQ.COM
The CMU hand data does not have hand bounding box information. There's only hand keypoints coordinate. How do you guys generate the ymin, xmin, ymax, xmax ?
Is the bbox coordinates directly from the min, max value among 21 points?
@qqwinn要看我写的发电机？留个联系方式我发给你。：|）
上一个邮箱我貌似填错了，真是不好意思
~qqwinn@aliyun.com这次没错:: |Hello, now I also want to train the model of my own data set, could you share it?thaks
Is the bbox coordinates directly from the min, max value among 21 points?
Yes, expand some pixels will be fine.
@ldfinfontainebleau
Hello, I am a student of the orientation of human posture estimation, hoping to get your help about how do I train my own data set models，thanks，
@ldfinfontainebleau
Hello, I am a student of the orientation of human posture estimation, hoping to get your help about how do I train my own data set models，thanks，
Of course， u can send your questions to 15620271968@163.com， I will reply as soon as I see.
@qqwinn 要看我写的generator？留个联系方式我发给你。: |)
上一个邮箱我貌似填错了，真是不好意思～
qqwinn@aliyun.com 这次没错 ：：|没事的，这两天在忙一个项目，还没来得及发给你，晚上发给你。
可以发一份给我吗，谢谢 201923668@qq.com
可以发我一份吗？最好保存在一个txt，然后在这个问题回复上传，这样就不会有这么多人问你要了
我的邮箱是351391905@qq.com，谢谢你哦！
@qqwinn 要看我写的generator？留个联系方式我发给你。: |)
Hello
I want to train this project by my own model
but I don't know how to make the generator
so could you send me a generator file?
I want to show your generator
maskrider55575@gmail.com
你好： 这个是别人写的generator.py文件，后续有什么心得可以交流哈～～～ 王琦琦
…
------------------------------------------------------------------ 发件人：mathduanxu notifications@github.com 发送时间：2019年5月7日(星期二) 14:44 收件人：timctho/convolutional-pose-machines-tensorflow convolutional-pose-machines-tensorflow@noreply.github.com 抄 送：qqwinn qqWinn@aliyun.com; Mention mention@noreply.github.com 主 题：Re: [timctho/convolutional-pose-machines-tensorflow] About training data, What is bg_img_dir ? (#48) @ldfinfontainebleau 您好，我在生成数据的时候也遇到了同样的问题，方便的话，可否也将代码共享我一下呀？谢谢啦，邮箱：dx13386842300@gmail.com ,非常感谢！ — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.
hello
could you send me a copy file? (generator)
I want to study training model
maskrider55575@gmail.com
您好，可以求一份generator的分享吗？十分感谢，这里邮箱是：296269258@qq.com
分享我的，我这个代码里用GetData类代替generator，注意json文件读取过程就行，加了三种数据增强
run_training.txt
Hello, I have a question for you,How did MPII make the public dataset a data file for a.txt file
I write my generator and ignore bg_img, it is training now. Everything seems fine.
what dataset can be used in this repo?
Do you mean data of the training program? I used it on hand key points. My training data just have two parts :
1.hands' pictures * .jpg
2.keypoints *.json
You could write your generator with your own datasets. I think it will be fine. (xiaolian)thks, but how to get key point efficiently, Is it manually marked?
Yes, manually and precisely.
你好，我在做手势识别（只需要0-5一共6个手势）的论文，
标注一个图片要标记20多个骨骼点的坐标，请问有什么高效的标注方法吗？如果1000张图片，手动标注的话，需要的工作量太巨大。
可以发下你的标注数据的程序吗？
你好： 这个是别人写的generator.py文件，后续有什么心得可以交流哈～～～ 王琦琦
…
------------------------------------------------------------------ 发件人：mathduanxu notifications@github.com 发送时间：2019年5月7日(星期二) 14:44 收件人：timctho/convolutional-pose-machines-tensorflow convolutional-pose-machines-tensorflow@noreply.github.com 抄 送：qqwinn qqWinn@aliyun.com; Mention mention@noreply.github.com 主 题：Re: [timctho/convolutional-pose-machines-tensorflow] About training data, What is bg_img_dir ? (#48) @ldfinfontainebleau 您好，我在生成数据的时候也遇到了同样的问题，方便的话，可否也将代码共享我一下呀？谢谢啦，邮箱：dx13386842300@gmail.com ,非常感谢！ — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.hello
could you send me a copy file? (generator)
I want to study training model
maskrider55575@gmail.com
我创建了一个QQ群（754876877），讨论手势识别和姿态识别 ，我来创建一个
点击链接加入群聊【手势识别和姿态识别】：https://jq.qq.com/?_wv=1027&k=5JJ3e5F
分享我的，我这个代码里用GetData类代替generator，注意json文件读取过程就行，加了三种数据增强
run_training.txt
你好，我在做手势识别（只需要0-5一共6个手势）的论文，
标注一个图片要标记20多个骨骼点的坐标，请问有什么高效的标注方法吗？如果1000张图片，手动标注的话，需要的工作量太巨大。
可以发下你的标注数据的程序吗？
我创建了一个QQ群（754876877），讨论手势识别和姿态识别 ，我来创建一个
点击链接加入群聊【手势识别和姿态识别】：https://jq.qq.com/?_wv=1027&k=5JJ3e5F
my emali is 787940575@qq.com,i have some problems in generator.py,could anyone give me a generator.py? thank you very much in advance!
分享我的，我这个代码里用GetData类代替generator，注意json文件读取过程就行，加了三种数据增强
run_training.txt
After training, how we convert the model into pkl format? It always report can't find conv1_1 from the loaded weight file？ How can we test images ？ Can you share the test code？
分享我的，我这个代码里用GetData类代替generator，注意json文件读取过程就行，加了三种数据增强
run_training.txt
@huangshaoguang 训练好的模型，是如何加载进程序进行测试的，发现load_weights_from_file这个函数如何加载训练好的模型
@qqwinn 要看我写的generator？留个联系方式我发给你。: |)
Can I get a "generator"? (My Email : qhstj0123@gmail.com)
@qqwinn 要看我写的generator？留个联系方式我发给你。: |)
上一个邮箱我貌似填错了，真是不好意思～
qqwinn@aliyun.com 这次没错 ：：|可以给我转发一份吗？谢谢
@qqwinn 要看我写的generator？留个联系方式我发给你。: |)
上一个邮箱我貌似填错了，真是不好意思～
qqwinn@aliyun.com 这次没错 ：：|
可以给我转发一下吗？1055029796@qq.com，谢谢
@qqwinn 要看我写的generator？留个联系方式我发给你。: |)
您好，可以发我份吗？非常感谢 ：-）
zhanghay163@163.com