def analyse_questions(video_dicts):
counts = {}
for video in video_dicts:
question_types = video["question_types"]
for question in question_types:
increment_in_map_(counts, question)
num_questions = sum([cnt for _, cnt in counts.items()])
counts = counts.items()
counts = sorted(counts, key=lambda pair: pair[0])
print(f"{'Question Type' :<20}{'Occurrences' :<15}Frequency")
for question_type, count in counts:
print(
f"{question_type:<20}{count:<15}{(count / num_questions) * 100:.3}%"
)
print(f"Total number of questions: {num_questions}")
q_types = [q_cnt[0] + 1 for q_cnt in counts]
cnts = [q_cnt[1] for q_cnt in counts]
fontsize = 18
plt.figure(figsize=(12, 6))
plt.bar(q_types, cnts)
plt.xlabel("Question Type", fontsize=fontsize)
plt.ylabel("Number of questions", fontsize=fontsize)
plt.xticks(q_types, fontsize=fontsize)
plt.yticks([0, 400, 800, 1200, 1600], fontsize=fontsize)
plt.show()
def _analyse_q_6(qa_pairs):
print("\nAnalysing state transition QA pairs...")
action_cnt = {}
for question, answer in qa_pairs:
increment_in_map_(action_cnt, answer)
_print_cnt_dict(action_cnt, "Action")
def _analyse_q_5(qa_pairs):
print("\nAnalysing repeating action QA pairs...")
event_cnt = {}
for question, answer in qa_pairs:
increment_in_map_(event_cnt, answer)
_print_cnt_dict(event_cnt, "Event")
def _analyse_q_4(qa_pairs):
print("\nAnalysing repetition count QA pairs...")
event_cnt = {}
for question, answer in qa_pairs:
event = question.split(" ")[-1][:-1]
increment_in_map_(event_cnt, event)
_print_cnt_dict(event_cnt, "Event")
def _analyse_q_3(qa_pairs):
print("\nAnalysing prop changed QA pairs...")
prop_cnt = {}
for question, answer in qa_pairs:
prop = answer.split(" ")[1]
increment_in_map_(prop_cnt, prop)
_print_cnt_dict(prop_cnt, "Property")
def _analyse_q_8(qa_pairs):
print("\nAnalysing counterfactual QA pairs...")
rock_colours = {}
answers = {}
for question, answer in qa_pairs:
splits = question.split(" ")
rock_colour = splits[-2]
increment_in_map_(rock_colours, rock_colour)
increment_in_map_(answers, answer)
_print_cnt_dict(rock_colours, "Rock colour")
_print_cnt_dict(answers, "Answer")
def _analyse_q_7(qa_pairs):
print("\nAnalysing explanation QA pairs...")
rot_cnt = {}
ans_cnt = {}
for question, answer in qa_pairs:
splits = question.split(" ")
rotation = splits[3]
increment_in_map_(rot_cnt, rotation)
increment_in_map_(ans_cnt, answer)
_print_cnt_dict(rot_cnt, "Rotation")
_print_cnt_dict(ans_cnt, "Answer")
def _analyse_q_0(qa_pairs):
print("\nAnalysing property QA pairs...")
prop_cnt = {}
prop_val_cnt = {"colour": {}, "rotation": {}}
for question, answer in qa_pairs:
prop = question.split(" ")[1]
prop_val = answer
increment_in_map_(prop_cnt, prop)
increment_in_map_(prop_val_cnt[prop], prop_val)
_print_cnt_dict(prop_cnt, "Property")
_print_cnt_dict(prop_val_cnt["colour"], "Colour Value")
_print_cnt_dict(prop_val_cnt["rotation"], "Rotation Value")
def count_rotations(video_dicts):
rotations = {}
num_frames = 0
for video in video_dicts:
frames = video["frames"]
for frame in frames:
num_frames += 1
objects = frame["objects"]
for obj in objects:
if obj["class"] == "octopus":
increment_in_map_(rotations, obj["rotation"])
print(f"{'Octopus rotations' :<20}{'Occurrences' :<15}Frequency")
for rotation, num in rotations.items():
print(f"{rotation:<20}{num:<15}{(num / num_frames) * 100:.3g}%")
print(f"\nTotal number of frames: {num_frames}\n")
def count_events(video_dicts):
event_list = []
num_frame_changes = 0
for video in video_dicts:
events = video["events"]
num_frame_changes += len(events)
flat_events = [event for subevents in events for event in subevents]
events = [extract_event(event) for event in flat_events]
event_list.extend(events)
event_dict = {}
for event in event_list:
increment_in_map_(event_dict, event)
print(f"{'Event name' :<20}{'Occurrences' :<15}Frequency")
for event, num in event_dict.items():
print(f"{event:<20}{num:<15}{(num / num_frame_changes) * 100:.3g}%")
print(f"\nTotal number of frame changes: {num_frame_changes}\n")
def count_fish_eaten(video_dicts):
fish_eaten = {}
num_videos = 0
for video in video_dicts:
events = video["events"]
num_videos += 1
num_fish_eaten = 0
for event_list in events:
if "eat fish" in event_list:
num_fish_eaten += 1
increment_in_map_(fish_eaten, num_fish_eaten)
fish_eaten = fish_eaten.items()
fish_eaten = sorted(fish_eaten, key=lambda eaten: eaten[0])
print(f"{'Fish eaten' :<15}{'Occurrences' :<15}Frequency")
for num_eaten, cnt in fish_eaten:
print(f"{num_eaten:<15}{cnt:<15}{(cnt / num_videos) * 100:.3g}%")
print(f"Total number of videos: {num_videos}")
def _gen_repeating_action_question(self):
"""
Generate a question asking about which event occurs a given number of times
Q: What does the <object> do <n> times?
A: <event>
Note: The object will always be the octopus
:return: (question: str, answer: str)
"""
event_counts = self._count_events()
if event_counts.get(NO_EVENT) is not None:
del event_counts[NO_EVENT]
events = list(event_counts.keys())
random.shuffle(events)
# Track the number of times each count occurs
counts = {}
for _, count in event_counts.items():
util.increment_in_map_(counts, count)
question_count = None
question_event = None
for event in events:
count = event_counts[event]
num_counts = counts[count]
if count != 0 and num_counts == 1:
question_count = count
question_event = event
if question_count is None:
return None
question = f"What does the octopus do {question_count} times?"
answer = question_event
return question, answer
def count_colours(video_dicts):
rock_colours = {}
octo_colours = {}
num_frames = 0
for video in video_dicts:
frames = video["frames"]
for frame in frames:
num_frames += 1
objects = frame["objects"]
for obj in objects:
if obj["class"] == "rock":
increment_in_map_(rock_colours, obj["colour"])
elif obj["class"] == "octopus":
increment_in_map_(octo_colours, obj["colour"])
print(f"{'Rock colour' :<20}{'Occurrences' :<15}Frequency")
for colour, num in rock_colours.items():
print(f"{colour:<20}{num:<15}{(num / num_frames) * 100:.3g}%")
print(f"\n{'Octopus colour' :<20}{'Occurrences' :<15}Frequency")
for colour, num in octo_colours.items():
print(f"{colour:<20}{num:<15}{(num / num_frames) * 100:.3g}%")
print(f"\nTotal number of frames: {num_frames}\n")
def _analyse_q_1(qa_pairs):
print("\nAnalysing relation QA pairs...")
rel_cnt = {}
ans_cnt = {}
for question, answer in qa_pairs:
words = question.split(" ")
if words[3] in ["close", "above", "below"]:
increment_in_map_(rel_cnt, words[3])
else:
increment_in_map_(rel_cnt, words[4])
increment_in_map_(ans_cnt, answer)
_print_cnt_dict(ans_cnt, "Answer")
_print_cnt_dict(rel_cnt, "Relation")