search

xixiareone / counting-pigs

GNU General Public License v3.0
45 stars 12 forks source link

Confused when reading h5 file #3

Open MarvelousV opened 4 years ago

MarvelousV commented 4 years ago

Hi, I just downloaded the datasets. When I read the file, I found that the 72x72 image contains 9 smaller images, the paper suggested that each large image will be cropped into 72x72 images but not mentioned that it will be further cropped or augmented. Is this expected behavior for the dataset? I orginally thought it would one 72x72 image.

image

Also the label for each image is 18x18, and I am confused because if the image is divided into 9 small ones, then the label should be the same size as 24x24.

image

Could you please share some details or something I miss while reading the paper?

MarvelousV commented 4 years ago

Also could you please kindly provide the image datasets? That would be really helpful!

xixiareone commented 4 years ago

Thank you very much for your attention to our work, we provide the original image, data expansion and data resize in the provided code, we are processing once and then directly feed the model, and then 72 downsampling twice to get 18, the size and the original annotation size is corresponding.

------------------ 原始邮件 ------------------ 发件人: "Tiecheng Wang"<notifications@github.com>; 发送时间: 2020年7月8日(星期三) 凌晨5:15 收件人: "xixiareone/counting-pigs"<counting-pigs@noreply.github.com>; 抄送: "Subscribed"<subscribed@noreply.github.com>; 主题: [xixiareone/counting-pigs] Confused when reading h5 file (#3)

Hi, I just downloaded the datasets. When I read the file, I found that the 72x72 image contains 9 smaller images, the paper suggested that each large image will be cropped into 72x72 images but not mentioned that it will be further cropped or augmented. Is this expected behavior for the dataset? I orginally thought it would one 72x72 image.

Also the label for each image is 18x18, and I am confused because if the image is divided into 9 small ones, then the label should be the same size as 24x24.

Could you please share some details or something I miss while reading the paper?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

MarvelousV commented 4 years ago

Hi, thanks for your quick reply! I only found h5 files from the Baidu link and can not find any images in jpg/png or other formats. And I noticed from previous issues that some asked for image datasets and those links are expired now. Could you provide the images or a link to the dataset if possible?

I still can not understand why there are 9 images in one 72x72 image. Since you mentioned "sliding the windows" in Chapter 3.3 of the paper, then the 9 images should be slightly different and the label should also be different, but for each 72x72 image, there is only one 18x18 label, could you please further explain this? Thank you in advance!

xixiareone commented 4 years ago

链接: https://pan.baidu.com/s/1OWAWMQmpw175hKzfTys0kw 提取码: r89a This is the original image data set, and we process the data in the manner of this code. For more details, you can look at this code https://github.com/gramuah/ccnn&nbsp;and process it according to his requirements.

------------------ 原始邮件 ------------------ 发件人: "Tiecheng Wang"<notifications@github.com>; 发送时间: 2020年7月8日(星期三) 中午11:28 收件人: "xixiareone/counting-pigs"<counting-pigs@noreply.github.com>; 抄送: "Gin"<xixiareone@foxmail.com>;"Comment"<comment@noreply.github.com>; 主题: Re: [xixiareone/counting-pigs] Confused when reading h5 file (#3)

Hi, thanks for your quick reply! I only found h5 files from the Baidu link and can not find any images in jpg/png or other formats. And I noticed from previous issues that some asked for image datasets and those links are expired now. Could you provide the images or a link to the dataset if possible?

I still can not understand why there are 9 images in one 72x72 image. Since you mentioned "sliding the windows" in Chapter 3.3 of the paper, then the 9 images should be slightly different and the label should also be different, but for each 72x72 image, there is only one 18x18 label, could you please further explain this? Thank you in advance!

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

MarvelousV commented 4 years ago

Thanks a lot!

MarvelousV commented 4 years ago

The real-life images are not allowed to share, right?

xiaofei-du commented 4 years ago

链接: https://pan.baidu.com/s/1OWAWMQmpw175hKzfTys0kw 提取码: r89a This is the original image data set, and we process the data in the manner of this code. For more details, you can look at this code https://github.com/gramuah/ccnn&nbsp;and process it according to his requirements.

Could you please reshare the link? i think it expired. Thanks!