Can't get the results of yes/no, number and other question.

[wanmeiziyoulan]

Hi, I have downloaded the vqa data (the link has been provided) and run the training. But the pkl file deserialized are some arrays, so I don't known how to distinguish the type of yes/no, number and other. How can I get the result of specific type of question?

Thank you,
WeimingZhang

[zcyang]

Hi,

I just pushed a script data_vqa/preprocess_data.py which I used to preprocess the data. You can rerun it to get the question type information.

best,
zichao

[wanmeiziyoulan]

@zcyang Thanks a lot!

[mikirui]

@zcyang
And I find that there are some problems in preprocess_data.py(line 67-80):

for idx, q_id in enumerate(train_question_ids):
    question = vqa.loadQuestion(q_id)[0]
    question = process_sentence(question)
    question = question.split()
    for word in question:
        question_dict_count[word] = question_dict_count.get(word, 0) + 1
    answer = vqa.loadAnswer(q_id)[0]
    answer_new = [process_answer(ans) for ans in answer]
    for word in answer_new:
        answer_dict_count[word] = answer_dict_count.get(word, 0) + 1
    train_questions.append(question)
    train_answers.append(answer)
    if idx % 1000 == 0:
        print 'finished processing %d in train' %(idx)

there are no methods like vqa.loadQuestion, vqa.loadAnswer in vqa tools: https://github.com/VT-vision-lab/VQA, are there some other codes for this?

And there are 10 answers for a question in VQA dataset, so do you just select one answer randomly or get all 10 answers as labels during training process for both open-ended and multiple choice task?

Looking for your reply, thanks~

[mikirui]

@marielladimiccoli
Sorry, I can't find the codes for it. If you just want to preprocess the VQA dataset, I think https://github.com/jiasenlu/HieCoAttenVQA/tree/master/prepro may be good for you, but the processed data may not have the same format as this project.

[marielladimiccoli]

Hi, Thank you ! I actually need just to execute the preprocess_data.py of @zcyang, <https://github.com/zcyang> finally could you do it? so I need the methods loadQuestion and LoadAnswer() for that, How did you do?

…

________________________________ Da: mikirui <[email protected]> Inviato: lunedì 9 gennaio 2017 16.06.48 A: zcyang/imageqa-san Cc: marielladimiccoli; Mention Oggetto: Re: [zcyang/imageqa-san] Can't get the results of yes/no, number and other question. (#6) @marielladimiccoli<https://github.com/marielladimiccoli> Sorry, I can't find the codes for it. If you just want to preprocess the VQA dataset, I think https://github.com/jiasenlu/HieCoAttenVQA/tree/master/prepro<url> may be good for you, but the processed data may not have the same format as this project. - You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#6 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/ARDZ5dgq_g8GkfgoLA6c4aYCaYLQgrzhks5rQk0IgaJpZM4K_sen>.

[mikirui]

@marielladimiccoli
Sorry, I didn't run the code finally... I just want to know how they process VQA dataset.
Maybe you can try to replace
question = vqa.loadQuestion(q_id)[0]
answer = vqa.loadAnswer(q_id)[0]
by
question = vqa.qqa[q_id]['question']
qa = vqa.loadQA(q_id)[0]
answer = qa['multiple_choice_answer']

But I haven't try the replaced code I mentioned, if there are some mistakes, I think you can check the API in https://github.com/VT-vision-lab/VQA/blob/master/PythonHelperTools/vqaTools/vqa.py and then check the structure of VQA dataset in http://visualqa.org/download.html