Json labels for the three VQA datsets

[tmp12316]

Thank you so much for your great work! Are there some official scripts to process VQA labels into conversation format for the 3 VQA datasets? Or could you please provide the pre-processed json files of them? Thanks.

[LiangXin1001]

Thank you so much for your great work! I also encountered this issue. Although I can download all the images, I still don't know how to generate the contents for the JSON files, such as test.json and train.json. Moreover, in the files from Hugging Face you provided, I found that the questions are all of a type that only requires a 'yes' as an answer. Can such questions be included as content in the JSON files?

[LiangXin1001]

And also, when I input the model_vqa_med.py code to GPT-4, it suggested that the input JSON file should look like this: { "id": "unique_question_identifier", "conversations": [ { "from": "user", "value": "text question" }, { "from": "gpt", "value": "expected answer" } ], "image": "optional_image_path.jpg" } Is this correct?

[Eldo-rado]

Thank you so much for your great work! I also encountered this issue. Although I can download all the images, I still don't know how to generate the contents for the JSON files, such as test.json and train.json. Moreover, in the files from Hugging Face you provided, I found that the questions are all of a type that only requires a 'yes' as an answer. Can such questions be included as content in the JSON files?

Hi, I meet the same issue, have you solved it? 😊

[wangpengyu-debug]

is there any differences when processing the data to { "id": "unique_question_identifier", "conversations": [ { "from": "user", "value": "text question" }, { "from": "gpt", "value": "expected answer" } ], "image": "optional_image_path.jpg" } this format?