search

ZrrSkywalker / MAVIS

Mathematical Visual Instruction Tuning for Multi-modal Large Language Models
MIT License
109 stars 1 forks source link

Missing Files in the MAVIS Dataset (Existing_Dataset_Augment) #3

Open ZS123-lang opened 1 month ago

ZS123-lang commented 1 month ago

Hi Authors,

Thank you for your contribution to the MAVIS dataset, which is an excellent resource for research in math vision.

However, I encountered an issue after downloading the dataset via the Google Drive link. It seems there are a large number of missing files in the folder Existing_Dataset_Augment where RuleBaseGeo_For_Vision_Dominant/depth//*_text.jpg are missing. For example, the file RuleBaseGeo_For_Vision_Dominant/depth3/24981/24982_text.jpg is not found.

The total number of missing files appears to be close to 10,000 images. Could you please look into this issue or provide guidance on obtaining the missing files?

Thank you again for your contribution!

ZS123-lang commented 1 month ago

sorry, almost 100,000 images are missing.

Coobiw commented 1 month ago

I've also found this problem. This is because that some of the images are absolute paths. But it is difficult to transfer them into the correct one. - -

ZS123-lang commented 1 month ago

@Coobiw the absolute path is not big deal, we can change to correct one. but for the missing files, such as all RuleBaseGeo_For_Vision_Dominant/depth/_text.jpg are not here, we have nothing to do but wait the authors to response.

Coobiw commented 1 month ago

Soga~ Indeed, as you said. I find the one which cannot be transfered into correct one has the feature as you said.