Data studio is not showing the preview images

🤗Datasets
1

I have a directory structure like:

train
   input
      001.png
   reference
      001.png
   source
      001.png
   target
      001.png
      001.txt
   metadata.csv
validation
   input
      001.png
      001.txt
   reference
      001.png
   source
      001.png
   metadata.csv

and configured the splits in my README. So the content of the metadata.csv is shown in the Data Studio, as expected. But the images are only shown as the string containing the path and not as a preview of the image itself.

The CSV is structured like:

source_file_name,reference_file_name,input_file_name,target_file_name,text,text2,"Edit type",name,"reference format","image format","reference ethnicity","image ethnicity","Reference sex","image sex",multiperson
"source/eye_0001.png","reference/eye_0001.png","input/eye_0001.png","target/eye_0001.png","Change the eyes of the person, keeping the rest of the image the same.","Change the eyes, keeping the rest of the image the same.","eye","eye_0001","Close-up","portrait","European","European","woman","woman","no"

What am I doing wrong here? How can I fix it?

(The data is in stablellama/change_eye_face_head_person · Datasets at Hugging Face but it’s gated till I’m happy with the data setup)

1 Like
2

There are several ways to improve the usability of DatasetViewer, but the most reliable method is to load and save or upload the dataset using the datasets library.

3

I’m just uploading files and not using any Python scripts here

1 Like
4

In cases where the dataset structure is simple, errors often do not occur, but it seems to be working intermittently. It may be better to raise an issue on DatasetViewer’s GitHub. I will ping the author just in case. @lhoestq