You are a legend

Most of downloading urls of images in ocr_vqa dataset are no longer available. Everyone has to rerun the downloading script to get a small portion of ocr_vqa images in the LLaVA-v1.5-665k instruction dataset. I zip all images from original release from https://huggingface.co/datasets/howard-hou/OCR-VQA into a zip file. Everyone can easily download it and unzip to their path of ./ocr_vqa/images

It's parquet, not jpg, can not use to train directly

Most of downloading urls of images in ocr_vqa dataset are no longer available. Everyone has to rerun the downloading script to get a small portion of ocr_vqa images in the LLaVA-v1.5-665k instruction dataset. I zip all images from original release from https://huggingface.co/datasets/howard-hou/OCR-VQA into a zip file. Everyone can easily download it and unzip to their path of ./ocr_vqa/images

It's parquet, not jpg, can not use to train directly

You misunderstand my pull request. Please check the changed readme file. The downloading link for the ocr_vqa images are https://huggingface.co/datasets/weizhiwang/llava_v15_instruction_images/resolve/main/ocr_vqa_images_llava_v15.zip?download=true. The mentioned link is the original release.

Most of downloading urls of images in ocr_vqa dataset are no longer available. Everyone has to rerun the downloading script to get a small portion of ocr_vqa images in the LLaVA-v1.5-665k instruction dataset. I zip all images from original release from https://huggingface.co/datasets/howard-hou/OCR-VQA into a zip file. Everyone can easily download it and unzip to their path of ./ocr_vqa/images

It's parquet, not jpg, can not use to train directly

You misunderstand my pull request. Please check the changed readme file. The downloading link for the ocr_vqa images are https://huggingface.co/datasets/weizhiwang/llava_v15_instruction_images/resolve/main/ocr_vqa_images_llava_v15.zip?download=true. The mentioned link is the original release.

Dude, You are a Legend

hero!

Life saver!

Super hero!

Most of downloading urls of images in ocr_vqa dataset are no longer available. Everyone has to rerun the downloading script to get a small portion of ocr_vqa images in the LLaVA-v1.5-665k instruction dataset. I zip all images from original release from https://huggingface.co/datasets/howard-hou/OCR-VQA into a zip file. Everyone can easily download it and unzip to their path of ./ocr_vqa/images

It's parquet, not jpg, can not use to train directly

You misunderstand my pull request. Please check the changed readme file. The downloading link for the ocr_vqa images are https://huggingface.co/datasets/weizhiwang/llava_v15_instruction_images/resolve/main/ocr_vqa_images_llava_v15.zip?download=true. The mentioned link is the original release.

@Victorwz 您好，请问这个数据现在没法下载了吗？

你点一下链接应该直接可以下载的，我刚点了一下没问题

Thanks a lot!

Thanks buddy!

Hi, I am trying to use the LLaVA-v1.5-665k OCR-VQA dataset for training. I have downloaded the image zip from your link:

https://huggingface.co/datasets/weizhiwang/llava_v15_instruction_images/resolve/main/ocr_vqa_images_llava_v15.zip?download=true.

However, I do not have the corresponding JSON annotations that map each image to its questions and answers. Without this JSON, I cannot use the dataset for training or evaluation.
Could you please provide the JSON file for OCR-VQA in LLaVA-v1.5-665k or guide me on how to generate it from the existing resources?