File size: 5,254 Bytes
b0c0df0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 |
import argparse
import json
import os
import pdb
import re
from copy import deepcopy
from pathlib import Path
import numpy as np
# read json files
def read_json(path):
with open(path, "r") as fin:
datas = json.load(fin)
return datas
def write_json(path, data):
with open(path, "w") as fout:
json.dump(data, fout)
print("The format file has been saved at:{}".format(path))
return
def extract_time(paragraph):
prompt = "A specific example is : 20.8 - 30.0 seconds".lower()
paragraph = paragraph.lower().replace(prompt, "").replace("to", "-")
# Split text into sentences based on common delimiters
sentences = re.split(r"[!?\n]", paragraph)
# Keywords that might indicate the presence of time information
keywords = ["starts", "ends", "happens in", "start time", "end time", "start", "end", "happen"]
# filter sentences by keywords
candidates = []
for sentence in sentences:
# If sentence contains one of the keywords
if any(keyword in sentence for keyword in keywords):
candidates.append(sentence)
timestamps = []
# Compile patterns once outside the loop for better performance
main_pattern = re.compile(r"(\d+(?:\.\d+)?)\s*[–-]\s*(\d+(?:\.\d+)?)")
time_number_pattern = re.compile(r"\b(\d+(?:\.\d+)?)\b")
time_format_pattern = re.compile(r"\b(\d{1,2}:\d{2}:\d{2})\b")
fallback_pattern = re.compile(r"(\d+(?:\.\d+)?)\s*s?\s*[–-]\s*(\d+(?:\.\d+)?)\s*s?")
if len(sentences) == 0:
return []
# Check for The given query happens in m - n (seconds)
# Search only in candidates, not the entire paragraph
for sentence in candidates:
time_matches = main_pattern.findall(sentence)
if time_matches:
timestamps = [[float(start), float(end)] for start, end in time_matches]
break
# check for other formats e.g.:
# 1 .Starting time: 0.8 seconds
# Ending time: 1.1 seconds
# 2. The start time for this event is 0 seconds, and the end time is 12 seconds.
if len(timestamps) == 0:
times = []
for sentence in candidates:
time = time_number_pattern.findall(sentence)
if time:
time_in_sec = float(time[0])
times.append(time_in_sec)
times = times[: len(times) // 2 * 2]
timestamps = [(times[i], times[i + 1]) for i in range(0, len(times), 2)]
# Check for examples like:
# 3. The event 'person flipped the light switch near the door' starts at 00:00:18 and ends at 00:00:23.
if len(timestamps) == 0:
times = []
for sentence in candidates:
time = time_format_pattern.findall(sentence)
if time:
t = time[0]
else:
continue
# If time is in HH:MM:SS format, convert to seconds
if t.count(":") == 2:
h, m, s = map(int, t.split(":"))
time_in_sec = h * 3600 + m * 60 + s
elif t.count(":") == 1:
m, s = map(int, t.split(":"))
time_in_sec = m * 60 + s
times.append(time_in_sec)
times = times[: len(times) // 2 * 2]
timestamps = [(times[i], times[i + 1]) for i in range(0, len(times), 2)]
# Fallback: if no timestamps found, search for any two number patterns with dash
if len(timestamps) == 0:
# More comprehensive pattern to match various formats like:
# xx - xx, x.xx s - x.xx s, x.xxs - x.xxs, etc.
# Also handle en dash (–) and regular dash (-)
for sentence in candidates:
fallback_matches = fallback_pattern.findall(sentence)
if fallback_matches:
timestamps = [[float(start), float(end)] for start, end in fallback_matches]
break
results = []
for start, end in timestamps:
if end > start:
results.append([start, end])
else:
results.append([end, start])
if len(results) > 1:
results = results[:1]
return results
def iou(A, B):
max0 = max((A[0]), (B[0]))
min0 = min((A[0]), (B[0]))
max1 = max((A[1]), (B[1]))
min1 = min((A[1]), (B[1]))
return max(min1 - max0, 0) / (max1 - min0)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-f", default="your_result.json")
args = parser.parse_args()
datas = read_json(args.f)
num = len(datas)
# miou
ious = []
for k in datas.keys():
vid, caption, gt = k.split(">>>")
pred = datas[k]
gt = eval(gt)
timestamps = extract_time(pred)
if len(timestamps) != 1:
print(f"pred={pred},timestamps={timestamps}")
timestamps = [[gt[1] + 10, gt[1] + 20]]
# print(f"GT: {gt}, Pred: {timestamps[0]}")
ious.append(iou(gt, timestamps[0]))
Result = {0.3: 0, 0.5: 0, 0.7: 0}
for c_iou in [0.3, 0.5, 0.7]:
for cur_iou in ious:
if cur_iou >= c_iou:
Result[c_iou] = Result[c_iou] + 1
print("IOU 0.3: {0}\nIOU 0.5: {1}\nIOU 0.7: {2}\nmIOU".format(Result[0.3] * 100 / num, Result[0.5] * 100 / num, Result[0.7] * 100 / num), sum(ious) * 100 / num)
|