Create get_dataset_stat.py
Browse files- get_dataset_stat.py +97 -0
get_dataset_stat.py
ADDED
|
@@ -0,0 +1,97 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from datasets import load_dataset
|
| 2 |
+
|
| 3 |
+
# Load the dataset
|
| 4 |
+
dataset = load_dataset("LVHaystack/LongVideoHaystack")
|
| 5 |
+
|
| 6 |
+
def calculate_statistics(dataset, splits=['train', 'test', 'val', 'test_tiny']):
|
| 7 |
+
statistics = {
|
| 8 |
+
"total_videos": 0,
|
| 9 |
+
"total_clips": 0,
|
| 10 |
+
"total_video_duration_hours": 0,
|
| 11 |
+
"total_vclip_duration_hours": 0,
|
| 12 |
+
"total_frames": 0,
|
| 13 |
+
"total_keyframes": 0,
|
| 14 |
+
"total_qa_pairs": 0,
|
| 15 |
+
"average_frames_per_video": 0,
|
| 16 |
+
"average_keyframes_per_question": 0,
|
| 17 |
+
"average_duration_per_video": 0,
|
| 18 |
+
"average_duration_per_vclip": 0,
|
| 19 |
+
"total_clips_deduplicated": 0,
|
| 20 |
+
"total_video_duration_hours_dedup": 0,
|
| 21 |
+
}
|
| 22 |
+
|
| 23 |
+
# Sets for deduplication
|
| 24 |
+
unique_video_ids = set()
|
| 25 |
+
unique_clips = set()
|
| 26 |
+
|
| 27 |
+
# Variables for calculations
|
| 28 |
+
total_clips = 0
|
| 29 |
+
total_video_duration = {}
|
| 30 |
+
total_vclip_duration = {}
|
| 31 |
+
total_frames = {}
|
| 32 |
+
total_keyframes = 0
|
| 33 |
+
total_questions = 0
|
| 34 |
+
|
| 35 |
+
for split in splits:
|
| 36 |
+
for entry in dataset[split]:
|
| 37 |
+
video_id = entry['video_metadata']['video_id']
|
| 38 |
+
clip_id = entry['vclip_id']
|
| 39 |
+
|
| 40 |
+
# Count unique videos
|
| 41 |
+
unique_video_ids.add(video_id)
|
| 42 |
+
|
| 43 |
+
# Track deduplicated clips
|
| 44 |
+
unique_clips.add(clip_id)
|
| 45 |
+
|
| 46 |
+
# Add clip duration to total duration
|
| 47 |
+
total_video_duration.update({entry['video_metadata']['video_id']: entry['video_metadata']['video_duration']})
|
| 48 |
+
|
| 49 |
+
# Add clip duration to total duration
|
| 50 |
+
total_vclip_duration.update({entry['vclip_id']: entry['video_metadata']['vclip_duration']})
|
| 51 |
+
|
| 52 |
+
# Add frame count to total frames
|
| 53 |
+
total_frames.update({entry['video_metadata']['video_id']: entry['video_metadata']['video_frame_count']})
|
| 54 |
+
|
| 55 |
+
# Add keyframes to total keyframes
|
| 56 |
+
total_keyframes += len(entry['frame_indexes_vclip'])
|
| 57 |
+
|
| 58 |
+
# Add question to total QA pairs
|
| 59 |
+
total_questions += 1
|
| 60 |
+
|
| 61 |
+
# Count video clips
|
| 62 |
+
total_clips += 1
|
| 63 |
+
|
| 64 |
+
# Populate statistics dictionary
|
| 65 |
+
statistics["total_videos"] = len(unique_video_ids)
|
| 66 |
+
statistics["total_clips"] = len(unique_clips)
|
| 67 |
+
statistics["total_video_duration_hours"] = sum(total_video_duration.values()) / 3600
|
| 68 |
+
statistics["total_vclip_duration_hours"] = sum(total_vclip_duration.values()) / 3600
|
| 69 |
+
statistics["total_frames"] = sum(total_frames.values())
|
| 70 |
+
statistics["total_keyframes"] = total_keyframes
|
| 71 |
+
statistics["total_qa_pairs"] = total_questions
|
| 72 |
+
statistics["average_frames_per_video"] = statistics["total_frames"] / len(unique_video_ids) if unique_video_ids else 0
|
| 73 |
+
statistics["average_keyframes_per_question"] = statistics["total_keyframes"] / total_questions if total_questions else 0
|
| 74 |
+
statistics["average_duration_per_video"] = statistics["total_video_duration_hours"] / len(unique_video_ids) * 3600 if unique_video_ids else 0
|
| 75 |
+
statistics["average_duration_per_vclip"] = statistics["total_vclip_duration_hours"] / len(unique_clips) * 3600 if unique_clips else 0
|
| 76 |
+
|
| 77 |
+
return statistics
|
| 78 |
+
|
| 79 |
+
def display(statistics, head="Dataset Statistics Summary"):
|
| 80 |
+
print("\n" + "="*40)
|
| 81 |
+
print(f" {head}")
|
| 82 |
+
print("="*40)
|
| 83 |
+
print(f"Total Videos: {statistics['total_videos']:,}")
|
| 84 |
+
print(f"Total Clips: {statistics['total_clips']:,}")
|
| 85 |
+
print(f"Total Video Duration (hours): {statistics['total_video_duration_hours']:.2f}")
|
| 86 |
+
print(f"Total Video Clip Duration (hours): {statistics['total_vclip_duration_hours']:.2f}")
|
| 87 |
+
print(f"Total Frames: {statistics['total_frames']:,}")
|
| 88 |
+
print(f"Total Keyframes: {statistics['total_keyframes']:,}")
|
| 89 |
+
print(f"Total QA Pairs: {statistics['total_qa_pairs']:,}")
|
| 90 |
+
print(f"Average Frames per Video: {statistics['average_frames_per_video']:.2f}")
|
| 91 |
+
print(f"Average Keyframes per Question: {statistics['average_keyframes_per_question']:.2f}")
|
| 92 |
+
print(f"Average Duration per Video (seconds, deduplicated): {statistics['average_duration_per_video']:.2f}")
|
| 93 |
+
print(f"Average Duration per Video Clip (seconds, deduplicated): {statistics['average_duration_per_vclip']:.2f}")
|
| 94 |
+
print("="*40 + "\n")
|
| 95 |
+
|
| 96 |
+
statistics = calculate_statistics(dataset)
|
| 97 |
+
display(statistics)
|