The "training_script" and "train/test split files" can't download.

zhufengx / SRN_multilabel

143 stars 39 forks source link

The "training_script" and "train/test split files" can't download. #10

Closed gjd2017 closed 6 years ago

gjd2017 commented 6 years ago

Hi, The "training_script" and "train/test split files" can't download, the link is wrong? Think you for reply.

zhufengx commented 6 years ago

Hi, @gjd2017 The links are fine. You may have to have access to Google Drive to download the files.

gjd2017 commented 6 years ago

Yes，I know. I have ask my USA friend for help, he also can't download. can you sand it by email? Thanks.

1414253285@qq.com

From: zhufengx Date: 2018-02-26 09:41 To: zhufengx/SRN_multilabel CC: gjd2017; Mention Subject: Re: [zhufengx/SRN_multilabel] The "training_script" and "train/test split files" can't download. (#10) Hi, @gjd2017 The links are fine. You may have to have access to Google Drive to download the files. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

zhufengx commented 6 years ago

Yes，I know. I have ask my USA friend for help, he also can't download. can you sand it by email? Thanks.

Hi, @gjd2017 , I have just updated new download links on BaiduYun. You may check again.

gjd2017 commented 6 years ago

Thanks, you are so great. @zhufengx

gjd2017 commented 6 years ago

Hi, Can you updated " cropped images for each human bounding box can be downloaded heres "(WIDER-Attribute) links on BaiduYun@zhufengx，thanks.

gjd2017 commented 6 years ago

I have original images of WIDER-Attribute,but I don't have cropped images of WIDER-Attribute.

zhufengx commented 6 years ago

I have original images of WIDER-Attribute,but I don't have cropped images of WIDER-Attribute.

Hi, the package of "cropped images of WIDER-Attribute" dataset exceeds the size limit (namely, 4GB) of BaiduYun, so that I cannot successfully upload it. While, you can easily get the cropped images using bounding box annotations of WIDER-Attribute dataset.

gjd2017 commented 6 years ago

Hi, zhufengx, Each image has so many bounding box(or target) annotations of WIDER-Attribute datase, which bounding box do you selected to crop? while, there are only one box in the "wider_att_train_imglist" ,such as "./Image_cropped/train/0--Parade/0_Parade_marchingband_1_385_11.jpg 3", and the "3" is means number of attribute?

zhufengx commented 6 years ago

Each image has so many bounding box(or target) annotations of WIDER-Attribute dataset, which bounding box do you selected to crop?

We have used all bounding boxes.

"./Image_cropped/train/0--Parade/0_Parade_marchingband_138511.jpg" corresponds to the 11-th bounding box of image "train/0--Parade/0_Parade_marchingband_1_385.jpg".
The label "3" means the number of positive attributes.
It's suggested that you crop the images and extract attribute labels all by yourself so that the extracted image crops and labels are exactly matched. It will not affect the train/test performance since we split images to train/test sets at image-level, but not crop-level.

gjd2017 commented 6 years ago

Oh，thank you very much, @zhufengx.