In the function tifa_score_single, there are these lines:
if question_answer_pair['question'] not in question_logs:
question_logs[question_answer_pair['question']] = question_answer_pair
choices=question_answer_pair['choices']
Consider changing this to:
if question_answer_pair['question'] not in question_logs:
question_logs[question_answer_pair['question']] = copy.deepcopy(question_answer_pair)
choices=question_answer_pair['choices']
Otherwise, whenever you run
result = tifa_score_single(vqa_model, filtered_questions, img_path)
you are changing the original filtered_questions, and result contain a reference to filtered_questions. Wierd things would happen. For example, if you make a new call with the same filtered_questions, the result from the previous call would be changed.
In the function
tifa_score_single, there are these lines:Consider changing this to:
Otherwise, whenever you run
result = tifa_score_single(vqa_model, filtered_questions, img_path)you are changing the originalfiltered_questions, and result contain a reference tofiltered_questions. Wierd things would happen. For example, if you make a new call with the samefiltered_questions, the result from the previous call would be changed.