File size: 12,955 Bytes
33569f9 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 | import json
import os
import datasets
import pandas as pd
def load_activitynet(split="test"):
data_root = "./dataset/activitynet"
data_path = f"{data_root}/annotations/sentence_temporal_grounding/{split}.json"
data = json.load(open(data_path))
qid, conv_data = 0, []
for video_id, meta_data in data.items():
video_path = None
for ext in ["mp4", "mkv", "webm"]:
tmp = os.path.join(f"{data_root}/videos", f"{video_id}.{ext}")
if os.path.exists(tmp):
video_path = tmp
break
assert video_path is not None
for i in range(len(meta_data["timestamps"])):
conv_data.append(
{
"video": video_path,
"duration": meta_data["duration"],
"timestamp": meta_data["timestamps"][i],
"sentence": meta_data["sentences"][i].strip(),
"qid": f"activitynet_{qid}",
}
)
qid += 1
return conv_data
def load_charades(split="test"):
data_root = "./dataset/charades"
data_path = f"{data_root}/Charades_anno/Charades_sta_{split}.json"
if not os.path.exists(data_path):
data = {}
old_data_path = f"{data_root}/Charades_anno/Charades_sta_{split}.txt"
data_csv = f"{data_root}/Charades_anno/Charades_v1_{split}.csv"
df = pd.read_csv(data_csv)
video_to_duration = dict(zip(df["id"], df["length"]))
for line in open(old_data_path):
if line.strip() == "":
continue
meta_data, sentence = line.split("##")
video_id, start, end = meta_data.split(" ")
if video_id not in data:
data[video_id] = {
"duration": video_to_duration[video_id],
"timestamps": [],
"sentences": [],
}
data[video_id]["timestamps"].append([float(start), float(end)])
data[video_id]["sentences"].append(sentence)
with open(data_path, "w") as f:
json.dump(data, f)
else:
data = json.load(open(data_path))
qid, conv_data = 0, []
for video_id, meta_data in data.items():
video_path = os.path.join(f"{data_root}/Charades_v1", f"{video_id}.mp4")
for i in range(len(meta_data["timestamps"])):
conv_data.append(
{
"video": video_path,
"duration": meta_data["duration"],
"timestamp": meta_data["timestamps"][i],
"sentence": meta_data["sentences"][i].strip(),
"qid": f"charades_{qid}",
}
)
qid += 1
return conv_data
def load_tvgbench_filter(split):
data_path = split
with open(data_path, "r", encoding="utf-8") as f:
data = json.load(f)
qid, conv_data = 0, []
for meta_data in data:
video = meta_data["video"]
duration = meta_data["duration"]
timestamps = meta_data["timestamp"]
sentences = meta_data["sentence"]
qid = meta_data["qid"]
pred = meta_data["pred"]
video_start = meta_data["video_start"]
video_end = meta_data["video_end"]
conv_data.append(
{
"video": video,
"duration": duration,
"timestamp": timestamps,
"pred": pred,
"sentence": sentences,
"qid": qid,
"video_start": video_start,
"video_end": video_end,
}
)
return conv_data
def load_tvgbench(split="default"):
"""
Load JSON data in TVGBench format.
Args:
data_path (str): Path to the JSON file in TVGBench format.
Returns:
list: A list containing processed data, where each element is a dictionary
in the format {'video': str, 'duration': float, 'timestamp': list[float, float], 'sentence': str, 'qid': str}.
Returns an empty list if the file does not exist or cannot be parsed.
"""
data_path = "./dataset/timer1/annotations/tvgbench.json"
with open(data_path, "r") as f:
raw_data = json.load(f)
qid_counter = 0
conv_data = []
for item in raw_data:
video_path = item["path"]
if not os.path.exists(video_path):
continue
duration_str = item["duration"]
answer_str = item["answer"]
question_str = item["question"]
start = item["start"]
end = item["end"]
duration = duration_str
parts = answer_str.split("-")
start_time = float(parts[0])
end_time = float(parts[1])
timestamp = [start_time, end_time]
sentence = question_str
if "source" in item and isinstance(item["source"], str):
source_filename = os.path.basename(item["source"])
source_prefix = (
os.path.splitext(source_filename)[0].replace(".", "_").replace("-", "_")
)
qid_str = f"{source_prefix}_{qid_counter}"
qid_counter += 1
conv_data.append(
{
"video": video_path,
"duration": duration,
"timestamp": timestamp,
"sentence": sentence,
"qid": qid_str,
"start": start,
"end": end,
}
)
return conv_data
def load_videomme(split="default"):
if split in ["test", "train"]:
split = "default"
assert split in ["short", "medium", "long", "default"]
data_root = "./dataset/videomme"
data_path = f"{data_root}/videomme"
conv_data = []
data = datasets.load_dataset(
"parquet", split="test", data_dir=data_path, streaming=True
)
for itm in data:
if split == "default" or itm["duration"] == split:
video_path = os.path.join(f"{data_root}/data", itm["videoID"] + ".mp4")
conv_data.append(
{
"video": video_path,
"question": itm["question"],
"options": [op[2:].strip() for op in itm["options"]],
"answer": ord(itm["answer"]) - ord("A"),
"duration": None,
"qid": f'videomme_{itm["question_id"]}',
}
)
return conv_data
def load_egoschema(split="default"):
if split in ["test", "train"]:
split = "default"
assert split in ["default", "subset"]
data_root = "./dataset/egoschema"
if split == "subset":
data_path = f"{data_root}/Subset"
else:
data_path = f"{data_root}/MC"
conv_data = []
data = datasets.load_dataset(
"parquet", split="test", data_dir=data_path, streaming=True
)
for itm in data:
video_path = os.path.join(f"{data_root}/videos", itm["video_idx"] + ".mp4")
conv_data.append(
{
"video": video_path,
"question": itm["question"],
"options": [op[2:].strip() for op in itm["option"]],
"answer": itm["answer"],
"duration": None,
"qid": f'egoschema_{itm["question_idx"]}',
}
)
return conv_data
def load_tempcompass(split="default"):
if split in ["test", "train", "default"]:
split = "multi-choice"
assert split in ["multi-choice", "captioning", "caption_matching", "yes_no"]
data_root = "./dataset/tempcompass"
data_path = f"{data_root}/questions/{split}.json"
conv_data = []
for key, value in json.load(open(data_path)).items():
video_path = os.path.join(f"{data_root}/videos", key + ".mp4")
for dim in value.keys():
for idx, itm in enumerate(value[dim]):
question, options, answer = itm["question"], [], itm["answer"]
if split == "yes_no":
options = ["yes", "no"]
answer = options.index(answer)
if split == "caption_matching":
tmp = question.split("\n")
question, options, answer = (
tmp[0],
[],
":".join(answer.split(":")[1:]).strip(),
)
for i in range(1, len(tmp)):
option = ":".join(tmp[i].split(":")[1:]).strip()
options.append(option)
answer = options.index(answer)
if split == "multi-choice":
tmp = question.split("\n")
question, options, answer = tmp[0], [], ord(answer[0]) - ord("A")
for i in range(1, len(tmp)):
options.append(tmp[i][2:].strip())
conv_data.append(
{
"video": video_path,
"question": question,
"options": options,
"answer": answer,
"duration": None,
"qid": f"tempcompass|{split}|{key}|{dim}|{idx}",
}
)
return conv_data
def load_mvbench(split="default"):
data_root = "./dataset/mvbench"
data_path = f"{data_root}/json"
DATASET_CONFIG = {
"action_sequence": f"{data_root}/video/star/Charades_v1_480/",
"action_prediction": f"{data_root}/video/star/Charades_v1_480/",
"action_antonym": f"{data_root}/video/ssv2_video/",
"fine_grained_action": f"{data_root}/video/Moments_in_Time_Raw/videos/",
"unexpected_action": f"{data_root}/video/FunQA_test/test/",
"object_existence": f"{data_root}/video/clevrer/video_validation/",
"object_interaction": f"{data_root}/video/star/Charades_v1_480/",
"object_shuffle": f"{data_root}/video/perception/videos/",
"moving_direction": f"{data_root}/video/clevrer/video_validation/",
"action_localization": f"{data_root}/video/sta/sta_video/",
"scene_transition": f"{data_root}/video/scene_qa/video/",
"action_count": f"{data_root}/video/perception/videos/",
"moving_count": f"{data_root}/video/clevrer/video_validation/",
"moving_attribute": f"{data_root}/video/clevrer/video_validation/",
"state_change": f"{data_root}/video/perception/videos/",
"fine_grained_pose": f"{data_root}/video/nturgbd/",
"character_order": f"{data_root}/video/perception/videos/",
"egocentric_navigation": f"{data_root}/video/vlnqa/",
"episodic_reasoning": f"{data_root}/video/tvqa/output_videos/",
"counterfactual_inference": f"{data_root}/video/clevrer/video_validation/",
}
conv_data = []
for file_name in os.listdir(data_path):
data_type = file_name.split(".")[0]
data = json.load(open(os.path.join(data_path, file_name)))
for qid, itm in enumerate(data):
video_name = itm["video"]
video_path = os.path.join(DATASET_CONFIG[data_type], video_name)
conv_data.append(
{
"video": video_path,
"question": itm["question"],
"options": itm["candidates"],
"answer": itm["candidates"].index(itm["answer"]),
"duration": None,
"qid": f"mvbench|{data_type}|{qid}",
}
)
if "start" in itm and "end" in itm:
video_name = (
itm["video"].split(".mp4")[0]
+ "_"
+ str(itm["start"]).replace(".", "-")
+ "_"
+ str(itm["end"]).replace(".", "-")
+ ".mp4"
)
video_path = os.path.join(
DATASET_CONFIG[data_type], "split", video_name
)
conv_data[-1]["video"] = video_path
else:
if "start" in itm:
conv_data[-1]["video_start"] = itm["start"]
if "end" in itm:
conv_data[-1]["video_end"] = itm["end"]
return conv_data
def _extract_qid(itm):
vtype, vid, question = (
None,
itm["video"].split("/")[-1].split(".")[0],
itm["sentence"],
)
video_path = itm["video"].lower()
if "cosmo" in video_path or "howto100m" in video_path:
vtype = "cosmo"
if "queryd" in video_path:
vtype = "queryd"
if "vtime" in video_path:
vtype = "internvid-vtime"
if ":" in vid:
vid = vid.split(":")[0][:-3]
if "didemo" in video_path:
vtype = "didemo"
if "yt_temporal_videos" in video_path:
vtype = "yt-temporal"
return f"my|{vtype}|{vid}|{question}"
|