search

miccunifi / SEARLE

[ICCV 2023] - Zero-shot Composed Image Retrieval with Textual Inversion
Other
151 stars 6 forks source link

Could you share the complete images of FashionIQ dataset ? #11

Closed Yang-bug-star closed 2 months ago

Yang-bug-star commented 2 months ago

In the official repository, only 74381 images can be downloaded

ABaldrati commented 2 months ago

Hi @Yang-bug-star

I think you can download the fashioniq dataset using the link in https://github.com/XiaoxiaoGuo/fashion-iq/issues/18 or using the following hugginface datasets repo https://huggingface.co/datasets/Plachta/FashionIQ/tree/main

Let me know if this solves your problem.

Yang-bug-star commented 2 months ago

But the link in XiaoxiaoGuo/fashion-iq#18 and  https://huggingface.co/datasets/Plachta/FashionIQ/tree/main  only contain 74381 images , which is not equal to the original number of images in fashioniq paper

------------------ 原始邮件 ------------------ 发件人: "miccunifi/SEARLE" @.>; 发送时间: 2024年8月27日(星期二) 下午4:11 @.>; @.**@.>; 主题: Re: [miccunifi/SEARLE] Could you share the complete images of FashionIQ dataset ? (Issue #11)

Hi @Yang-bug-star

I think you can download the fashioniq dataset using the link in XiaoxiaoGuo/fashion-iq#18 or using the following hugginface datasets repo https://huggingface.co/datasets/Plachta/FashionIQ/tree/main

Let me know if this solves your problem.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

ABaldrati commented 2 months ago

I don’t recall the exact figures from the FashionIQ paper, but there are a total of 74,381 FashionIQ images, so you should have access to all of them. Please note that some images fall into multiple categories, so the total number of images is not simply the sum of the images in each individual category.

Let me know if this solves your problem.

Yang-bug-star commented 2 months ago

By the way, I want to ask that did your reported results in paper are based on validation set of FashionIQ dataset or test set ? Because I see that the original test set doesn't contain ground truth

------------------ 原始邮件 ------------------ 发件人: "miccunifi/SEARLE" @.>; 发送时间: 2024年8月27日(星期二) 下午4:51 @.>; @.**@.>; 主题: Re: [miccunifi/SEARLE] Could you share the complete images of FashionIQ dataset ? (Issue #11)

I don’t recall the exact figures from the FashionIQ paper, but there are a total of 74,381 FashionIQ images, so you should have access to all of them. Please note that some images fall into multiple categories, so the total number of images is not simply the sum of the images in each individual category.

Let me know if this solves your problem.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

ABaldrati commented 2 months ago

As a standard practice, we report the results on the validation set of the FashionIQ