The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Error code: DatasetGenerationError
Exception: CastError
Message: Couldn't cast
video_path: string
data_type: string
duration: double
caption: string
qa_check_ratio: double
event_check_ratio: double
event_total: int64
event_list: struct<audio: list<item: struct<modal: string, type: string, character: string, content: string>>, v (... 91 chars omitted)
child 0, audio: list<item: struct<modal: string, type: string, character: string, content: string>>
child 0, item: struct<modal: string, type: string, character: string, content: string>
child 0, modal: string
child 1, type: string
child 2, character: string
child 3, content: string
child 1, visual: list<item: struct<modal: string, type: string, character: string, content: string>>
child 0, item: struct<modal: string, type: string, character: string, content: string>
child 0, modal: string
child 1, type: string
child 2, character: string
child 3, content: string
a2v_query_list: list<item: struct<modal: string, type: string, character: string, content: string, selected_visual_i (... 222 chars omitted)
child 0, item: struct<modal: string, type: string, character: string, content: string, selected_visual_indices: lis (... 210 chars omitted)
child 0, modal: string
child 1, type: string
child 2, character: string
child 3, content: string
child 4, selected_visual_indices: list<item: int64>
child 0, item: int64
child 5, visual_candidates_for_judge: list<item: struct<modal: string, type: string, character: string, content: string>>
child 0, item: struct<modal: string, type: string, character: string, content: string>
child 0, modal: string
child 1, type: string
child 2, character: string
child 3, content: string
child 6, selected_visual_indices_new: list<item: int64>
child 0, item: int64
child 7, window_range: list<item: int64>
child 0, item: int64
to
{'video_path': Value('string'), 'data_type': Value('string'), 'duration': Value('float64'), 'caption': Value('string'), 'qa_check_ratio': Value('float64'), 'event_check_ratio': Value('float64'), 'event_total': Value('int64')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1835, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 621, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
video_path: string
data_type: string
duration: double
caption: string
qa_check_ratio: double
event_check_ratio: double
event_total: int64
event_list: struct<audio: list<item: struct<modal: string, type: string, character: string, content: string>>, v (... 91 chars omitted)
child 0, audio: list<item: struct<modal: string, type: string, character: string, content: string>>
child 0, item: struct<modal: string, type: string, character: string, content: string>
child 0, modal: string
child 1, type: string
child 2, character: string
child 3, content: string
child 1, visual: list<item: struct<modal: string, type: string, character: string, content: string>>
child 0, item: struct<modal: string, type: string, character: string, content: string>
child 0, modal: string
child 1, type: string
child 2, character: string
child 3, content: string
a2v_query_list: list<item: struct<modal: string, type: string, character: string, content: string, selected_visual_i (... 222 chars omitted)
child 0, item: struct<modal: string, type: string, character: string, content: string, selected_visual_indices: lis (... 210 chars omitted)
child 0, modal: string
child 1, type: string
child 2, character: string
child 3, content: string
child 4, selected_visual_indices: list<item: int64>
child 0, item: int64
child 5, visual_candidates_for_judge: list<item: struct<modal: string, type: string, character: string, content: string>>
child 0, item: struct<modal: string, type: string, character: string, content: string>
child 0, modal: string
child 1, type: string
child 2, character: string
child 3, content: string
child 6, selected_visual_indices_new: list<item: int64>
child 0, item: int64
child 7, window_range: list<item: int64>
child 0, item: int64
to
{'video_path': Value('string'), 'data_type': Value('string'), 'duration': Value('float64'), 'caption': Value('string'), 'qa_check_ratio': Value('float64'), 'event_check_ratio': Value('float64'), 'event_total': Value('int64')}
because column names don't match
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1379, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 956, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1702, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1887, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
video_path string | data_type string | duration float64 | caption string | qa_check_ratio float64 | event_check_ratio float64 | event_total int64 |
|---|---|---|---|---|---|---|
finevideo_videos/_bWTNumXpTo.mp4 | video_with_audio | 23.28 | In a black and white, medium shot, a man in a dark, long-sleeved shirt stands on a stage, speaking passionately into a microphone he holds in his right hand against a backdrop of soft, inspirational, and slightly melancholic instrumental music. The background is a large, abstract, textured wall, with the heads of two g... | 0.88 | 1 | 5 |
finevideo_videos/tzssE66FJps.mp4 | video_with_audio | 20.766667 | A static, split-screen image of a galaxy is shown, diagonally divided into two contrasting views. As the image is displayed, a narrator with a calm, clear, and informative male voice explains the composition, stating, "This galaxy image is a diagonal combination of two images, one by Hubble on the upper left and the ot... | 0.88 | 1 | 3 |
finevideo_videos/12gB0JK9ueo.mp4 | video_with_audio | 34.067367 | A split-screen video is shown, where the top panel features a medium close-up of a press conference and the bottom panel displays a static wide shot of a large cargo ship, heavily loaded with shipping containers, that has collided with and significantly damaged a bridge. A "9 abc" logo with "WTVC CHATTANOOGA" written u... | 0.75 | 1 | 5 |
finevideo_videos/cVg0YCGVKtY.mp4 | video_with_audio | 49.16 | The audio opens with a low, ambient, and slightly mysterious electronic score as a male narrator begins speaking in a deep, serious, and inquisitive tone. Simultaneously, the video shows a medium shot of an astronaut from behind, standing on a desolate, rocky, gray surface under a dark, starry sky, wearing a white and ... | 1 | 1 | 8 |
finevideo_videos/uTjHb5yOtBQ.mp4 | video_with_audio | 66.166667 | A low-angle, shaky shot moves across a vast, snow-covered landscape where the low sun casts long, dark shadows from two bare tree trunks onto the pristine white snow, while the constant sound of strong, whistling wind creates a crisp, outdoor atmosphere. In the distance, a small, snow-covered stump stands on the right ... | 1 | 1 | 7 |
finevideo_videos/bdvzCeYmpRM.mp4 | video_with_audio | 43.12 | The audio opens with a gentle, ambient, and slightly mysterious electronic music track that evokes a sense of cosmic wonder as a close-up, animated shot shows a large, striped planet with green and white bands, similar to Jupiter. Small, blue and white spheres resembling Earths fall into its dark, open top, while a cir... | 1 | 1 | 10 |
finevideo_videos/QnB34NSY4W0.mp4 | video_with_audio | 70.503767 | In a medium close-up shot, a woman with long, wavy blonde hair wearing a red top with white polka dots speaks directly to the camera against a plain, light-colored wall with an out-of-focus shelf of framed pictures behind her. A small, white "Us" logo is in the top right corner as the camera remains static. With a seri... | 1 | 1 | 11 |
finevideo_videos/uti2mqFfsw0.mp4 | video_with_audio | 113.7136 | The audio opens with an upbeat, energetic electronic track with a driving beat as a man with reddish-brown hair and a beard, wearing a black t-shirt, stands in the center of the frame against a vibrant, digitally created image of a futuristic battle scene with robotic mechs and colorful fireworks. Gesturing with his ha... | 0.88 | 1 | 22 |
finevideo_videos/zKJnt11oHZ8.mp4 | video_with_audio | 32.9329 | In a medium close-up shot, a woman with long, wavy brown hair, wearing glasses, a maroon blazer over a black T-shirt with a purple hexagonal pattern, speaks directly to the camera against a dark, abstract background with warm, brownish-orange tones. As she speaks, the NASA Webb logo is visible in the top left corner wh... | 0.75 | 1 | 10 |
finevideo_videos/0lKH5Otczf8.mp4 | video_with_audio | 95.133333 | In a close-up, selfie-style shot from a slightly shaky, hand-held camera, a woman with dark, shoulder-length hair and a patterned top looks directly into the lens and speaks with an expressive face against a simple, light-colored wall. Simultaneously, the indistinct chatter and laughter of a crowd create a lively, bust... | 0.62 | 0.94 | 18 |
finevideo_videos/fIebZ4d1K_E.mp4 | video_with_audio | 64.3643 | A high-angle, wide shot captures a robot combat arena under green lighting, where a logo for "NHRL" is visible in the top left corner. The floor is a light gray, marbled surface, and the entire structure is enclosed by glass panels, behind which spectators watch from a dark event space with purple and green lighting. I... | 0.5 | 1 | 12 |
finevideo_videos/w4EWBXciVm8.mp4 | video_with_audio | 46.866667 | The video opens with a brief close-up of a woman with short, styled brown hair in a black judge's robe with a white lace collar, gesturing against a wood-paneled background, while on-screen text reads "POLYGAMY DEBATE PART 5." Simultaneously, a man begins speaking off-screen, and the scene immediately cuts to a close-u... | 0.75 | 1 | 16 |
finevideo_videos/HPQxPYWSdNQ.mp4 | video_with_audio | 54.366667 | A static, close-up shot shows a clear glass pill bottle filled with colorful capsules in shades of pink, purple, blue, and white, sitting on a reflective surface with more capsules scattered around its base against a blurry, out-of-focus cityscape illuminated by purple and pink lights. As this image is displayed with t... | 0.75 | 1 | 21 |
finevideo_videos/aejCbJGlOHU.mp4 | video_with_audio | 60.033333 | The video opens with a montage of a rider getting ready, beginning with a close-up of an orange t-shirt featuring a large, stylized "WIXC racing" logo, while the steady, rhythmic thumping of a heartbeat creates a sense of anticipation. The montage continues with quick cuts of the rider putting on a red and white jersey... | 0.75 | 0.86 | 14 |
finevideo_videos/MGQ9XvxxnH4.mp4 | video_with_audio | 54.187467 | A close-up, black-and-white shot focuses on the face of a man in a dimly lit environment where the background is completely dark, isolating him. The camera is shaky, creating a raw and intimate feel as it zooms in and out slightly, sometimes blurring before refocusing on his intense eyes. Simultaneously, the man on scr... | 0.88 | 1 | 8 |
finevideo_videos/kMIrgsvM9ZM.mp4 | video_with_audio | 59.066667 | A close-up, static shot shows a white, multi-tiered storage unit filled with a vast collection of colorful fidget toys, with the top shelf packed with rows of rainbow-colored items and the bottom shelf holding a jumble of various pop-it and spin-it toys. As a narrator with a high-pitched, extremely high-energy voice be... | 1 | 1 | 18 |
finevideo_videos/b8WMfsLoUAo.mp4 | video_with_audio | 94.385956 | A man with long, dark hair, a full beard, glasses, and a dark green hat over a plaid shirt and dark jacket enters the frame from the left, positioning himself in front of a large poster on a light-colored wall filled with text and colorful illustrations resembling a tarot card guide. As the stationary camera captures h... | 0.62 | 1 | 14 |
finevideo_videos/Om3bqoBQ00s.mp4 | video_with_audio | 24.491158 | A close-up shot features an anime-style girl with light brown hair styled in two long braids, each adorned with a small red flower. She has small, pointed ears, two small horns on her head, and her large, expressive eyes are a warm, golden-orange color as she looks directly at the viewer with a gentle and empathetic ex... | 1 | 1 | 5 |
finevideo_videos/Bg5iKXZMyTw.mp4 | video_with_audio | 58.825433 | In a close-up, selfie-style shot, a man with dark hair, a beard, and sunglasses, wearing a dark vest over a blue shirt, speaks directly to the camera while standing outdoors in a field of dry, harvested crops on a bright, sunny day. In a calm but urgent analytical tone, he states, "It seems quite clear that Ukraine at ... | 0.75 | 0.85 | 13 |
finevideo_videos/0zECzU_Ztt8.mp4 | video_with_audio | 62.062062 | The audio begins with a gentle, ambient electronic track that creates a thoughtful and slightly futuristic mood, as the video opens on two men in white lab coats in a laboratory. The man on the left, with short black hair, holds a green, textured object and gestures while the man on the right examines a prosthetic hand... | 0.88 | 1 | 11 |
finevideo_videos/clNDUoxzSIc.mp4 | video_with_audio | 182.43225 | The video opens with a black screen as a sharp, impactful sound effect plays, followed by a modern, electronic, and slightly tense musical score that creates a sense of anticipation. The screen transitions to a light gray background where a dark blue, stylized logo animates into view, showing a rounded rectangle with t... | 0.75 | 1 | 18 |
finevideo_videos/squSeBoQIaM.mp4 | video_with_audio | 41.9 | As a static, split-screen promotional image remains displayed against a black background, an off-screen male narrator speaks with an enthusiastic, clear, and fast-paced promotional tone. The visual is divided into two vertical panels: the left, on a dark green leafy background, advertises "RIOTSEEDS.COM" with a "COUPON... | 1 | 1 | 7 |
finevideo_videos/A5rp5JP9JsA.mp4 | video_with_audio | 99.432667 | As a calm and informative female narrator with a steady, clear voice begins to speak, a close-up shot shows a person's hands on the controls of a black mobility scooter. The left hand turns a silver knob labeled "SPEED," while the right hand rests near the key already in the ignition on a control panel that also featur... | 0.88 | 1 | 15 |
finevideo_videos/mlp2mQTEVtQ.mp4 | video_with_audio | 140.12 | The video opens on a light blue background with a bright, glowing orb at the center as green text appears, stating, "Plants have an immune system... and it's complicated." The scene transitions to an animation of a plant growing from a brown pot, while a female narrator with a calm, steady, and educational tone asks, "... | 1 | 1 | 21 |
finevideo_videos/uQ8F1eTNBcw.mp4 | video_with_audio | 42.166667 | The video opens with a close-up, selfie-style shot of two young men in a well-lit room with a window in the background. In the foreground, a man in a black barber's cape looks at the camera as a hairstylist, who is wearing a white baseball cap and a grey crushed-velvet shirt, stands behind him and runs his hands throug... | 1 | 1 | 9 |
finevideo_videos/3rBFwTZ2YvA.mp4 | video_with_audio | 59.976583 | A close-up, slightly shaky shot focuses on the cover of the book "AUTobiography of a YOGI" by Paramahansa Yogananda, which rests on a light-colored, textured surface and features a portrait of the author with long, dark hair, a nose ring, and a calm expression, wearing an orange-colored garment. Simultaneously, a male ... | 1 | 1 | 8 |
finevideo_videos/92SJ-zq2KBg.mp4 | video_with_audio | 115.682233 | The video opens on a static black screen with the words "Go Poop!" displayed in a simple, white, sans-serif font, as a narrator with a calm, conversational tone begins speaking, "Alright. Jazzy, ball. We're gonna play a little fetch." The scene cuts to a shaky, high-angle handheld shot of a tan and black dog sitting on... | 0.88 | 1 | 15 |
finevideo_videos/82S1CT1_vOw.mp4 | video_with_audio | 83.766667 | A close-up, handheld shot pans across a cluttered laboratory bench filled with complex scientific equipment, including a large glass Erlenmeyer flask, various tubes, and a network of wires connecting different machines. In the background, a black office chair sits to the left, and a wall of cabinets and more lab appara... | 0.88 | 1 | 11 |
finevideo_videos/4CuP9P-h7os.mp4 | video_with_audio | 46 | A woman with long, dark hair, wearing a bright red long-sleeved shirt, stands against a plain white background and smiles warmly at the camera, while a light, optimistic instrumental track plays. As she begins speaking with a clear and enthusiastic voice, asking, "Is the current pandemic affecting your income?", a 2D a... | 1 | 1 | 11 |
finevideo_videos/2cAATfXrOTo.mp4 | video_with_audio | 23.366667 | A close-up, top-down shot shows a white piece of paper on an art easel in a cluttered artist's workspace. At the top of the frame, a handwritten note reads, "I'm gonna draw another troll," with an arrow pointing down, and a comment box is superimposed on the canvas asking, "Do you decide about making a troll before you... | 0.75 | 1 | 7 |
finevideo_videos/I3S2Vv5kFic.mp4 | video_with_audio | 140.874067 | The video opens on a clean, white background with a small, abstract, golden, dotted shape at the top center, accompanied by a gentle, slightly melancholic, and hopeful instrumental piece that creates a serious, contemplative mood. Below the shape, a thin, multi-colored bar and the text "II GLOBAL PARLIAMENTARY SUMMIT" ... | 1 | 1 | 13 |
finevideo_videos/4PyTLRh7k5w.mp4 | video_with_audio | 39.53954 | The video opens with a close-up, slightly shaky shot of a young man with purple-dyed hair sitting at a table in what appears to be an outdoor restaurant patio, looking down at a black smartphone he holds in his hands. He is wearing a red and white t-shirt with a black graphic, while on-screen text at the top reads, "ST... | 0.62 | 0.83 | 6 |
finevideo_videos/G8WbG5GGsqM.mp4 | video_with_audio | 34.466667 | A close-up, selfie-style shot shows two young women sitting side-by-side in what appears to be the dimly lit interior of a vehicle with a warm, wooden panel behind them. The woman on the left, with long, straight brown hair with blonde highlights, wears a black t-shirt with pink lettering under a black jacket. The woma... | 0.88 | 0.83 | 6 |
finevideo_videos/5sYqH15qQaA.mp4 | video_with_audio | 57.566667 | A man with a beard, wearing a dark t-shirt, is centered in the frame looking directly at the camera in a room illuminated with blue light, featuring a string of blue LED lights on the ceiling. To his left is a list of roles for a murder mystery, and at the bottom is an Instagram logo next to the handle "@REALFLIXGAMES"... | 0.91 | 1 | 9 |
finevideo_videos/WkhgH3bvUp4.mp4 | video_with_audio | 192.966667 | The video opens with a static shot of a colorful, cartoon-style illustration of scientific items like beakers and microscopes, with "BioTech and science" in the center and a "PowerDirector" watermark in the bottom right, as a female narrator with a clear, calm, and educational voice begins, "Hello everyone. Welcome to ... | 0.88 | 1 | 26 |
finevideo_videos/u1r5-DzTPew.mp4 | video_with_audio | 158.125 | A static shot shows a logo on a plain white background. The logo features the text "NORTH TEXAS" in a bold, dark blue font with a light blue outline of the state of Texas between the words. To the left is the red United Methodist "Cross and Flame" logo, and below this, in a smaller, gray font, is the text "CONFERENCE O... | 0.75 | 0.95 | 19 |
finevideo_videos/Aj4Eg4QKKJs.mp4 | video_with_audio | 139 | A man with a beard, wearing a green t-shirt, a cross necklace, and a camouflage baseball cap with the word "stellar" on it, speaks directly to the camera in a close-up, selfie-style shot. He is outdoors on a bright, sunny day, with a clear blue sky and the green leaves of a tree visible in the background. In a calm, co... | 1 | 1 | 19 |
finevideo_videos/NlWPjAq9RXU.mp4 | video_with_audio | 27.3 | A static split-screen image is displayed against a dark, textured background with shimmering gold light effects, showing two photographs of Bruce Lee. The top panel is a color photograph of him lying in a casket wearing a light-colored shirt, while the bottom panel is a black-and-white photograph of him shirtless and s... | 0.75 | 1 | 3 |
finevideo_videos/bPrfVKW2fEc.mp4 | video_with_audio | 34.868167 | In a close-up shot, a man with long dark hair and a full beard, wearing a light blue shirt, rests his chin on his hand in a softly lit room with string lights visible behind him. As a large black text box appears over the center of the frame asking, "Is ChatGPT cheating?", the man simultaneously speaks directly to the ... | 0.75 | 1 | 7 |
finevideo_videos/jQGCL_JFD0w.mp4 | video_with_audio | 17.875 | In a medium close-up shot, a man with short, dark hair wearing a black t-shirt speaks and gestures, while a cutout image of a female nurse in blue scrubs is simultaneously visible in the foreground against a dark, out-of-focus background. With a clear, confident, and persuasive tone, he begins, "Nurses make great medic... | 0.75 | 1 | 6 |
finevideo_videos/tkBCwuA6cmk.mp4 | video_with_audio | 157.866667 | The video opens with a title card against a vibrant red background where, in the center, the text "GAMERZWORLD" is displayed in bold white letters, with the question "HOW MANY GAMES CAN YOU STORE ON THE XBOX SERIES S INTERNAL STORAGE" appearing below in a smaller white font; simultaneously, a dynamic, abstract animatio... | 0.88 | 0.93 | 15 |
finevideo_videos/-69fV0nLvDU.mp4 | video_with_audio | 70 | The video opens with a static shot of a gray, generic human head silhouette against a plain white background, while a series of quick, futuristic sound effects, including a high-pitched electronic whir, a sharp whoosh, and rapid percussive clicks, create a sense of digital transformation. As the silhouette turns black,... | 0.88 | 1 | 13 |
finevideo_videos/tL_osaNs9VY.mp4 | video_with_audio | 184.7 | The video opens with a multi-panel montage of a young boy with glasses, shown in various shirts and settings, including one with a pink "BINARY NUMBERS" poster, while a gentle, ambient electronic melody plays. Over this collage, the text "Johan Thomas Jiji" and "Fun Learning" appears in pink. The scene then transitions... | 0.88 | 1 | 19 |
finevideo_videos/0BlKX2Ic1yg.mp4 | video_with_audio | 59.433333 | In a dimly lit, wood-paneled room filled with antique furniture, including a lit table lamp and framed portraits, a man with dark hair wearing a dark three-piece suit with a patterned tie and pocket square sits in an ornate chair and looks off-camera to his left as he speaks with a calm, measured tone, stating, "Mr. Ma... | 0.75 | 0.95 | 20 |
finevideo_videos/eOGWj0ykCVU.mp4 | video_with_audio | 46.5 | A man with a black baseball cap worn backward and a grey hoodie speaks directly to the camera in a medium close-up shot, with a professional microphone visible in the foreground and a softly lit home kitchen in the background. In a friendly and enthusiastic tone, he says, "What's up, everyone? Welcome to the very first... | 0.88 | 0.92 | 12 |
finevideo_videos/rtkIs0RkrAc.mp4 | video_with_audio | 152.152 | The video opens with a medium close-up of a woman with long, wavy brown hair, wearing a grey top and a silver necklace, who is speaking directly to the camera against the backdrop of a well-lit room with a large, dark wood bookshelf. On a small table to her left, two framed pictures and a small white teddy bear are vis... | 1 | 1 | 21 |
finevideo_videos/o9ZKCtVVMng.mp4 | video_with_audio | 62.53 | The video begins with a static shot of a title card on a black background, which displays "Single Displacement Reaction" in large, bold, yellow text, with an "NCSSM Online" logo in the bottom right, while the audio is purely informational with no background music, creating a focused academic atmosphere. The scene trans... | 1 | 1 | 7 |
finevideo_videos/o-A8HMXq_5M.mp4 | video_with_audio | 126.266667 | The video opens on a medium shot of two animated food items, a hamburger on the right and a red soda cup on the left, sitting side-by-side on a table with a plate of french fries. Both have cartoon faces with large, expressive eyes and mouths, and the soda cup has eyelashes. The background is a softly blurred restauran... | 0.88 | 1 | 14 |
finevideo_videos/9k4K6Ckm-yw.mp4 | video_with_audio | 54.066667 | In a medium shot, a woman with graying hair and glasses, wearing a black off-the-shoulder top and black pants, sits in a wooden chair in front of a blue banner with the "GERALD R. FORD SCHOOL OF PUBLIC POLICY" text and a "M" logo repeated. In front of her, a small wooden table holds a vase of yellow and blue flowers an... | 1 | 1 | 13 |
finevideo_videos/4K2x4IVzrec.mp4 | video_with_audio | 119.933333 | The video opens with a very dark, close-up shot of a man in what appears to be a recording studio. He has long dreadlocks, a white baseball cap worn backward, dark sunglasses, a goatee, and is wearing a black t-shirt with colorful graphics. A professional microphone on a stand is visible in the foreground. Simultaneous... | 0.5 | 0.84 | 19 |
finevideo_videos/sy9w0LxXih8.mp4 | video_with_audio | 59.6 | A muscular man with a full, dark beard and a bald head stands in a room, looking directly at the camera. He is wearing a salmon-colored t-shirt with a black graphic of a split skull with arm-wrestling arms and the text "MONSTER" above it. In the background, a shelf filled with numerous gold trophies is visible, along w... | 0.88 | 1 | 11 |
finevideo_videos/Sgh-adBDT_o.mp4 | video_with_audio | 41.5 | In a static, medium close-up shot, a bald man in a black t-shirt speaks directly to the camera from what appears to be a home office. Behind him to his right, a white bookshelf is filled with books, including one with "THE PINEAPPLE THEORY" on its cover. A logo for "JAVELIN content management" is displayed at the top o... | 1 | 1 | 8 |
finevideo_videos/aYh4QfGdhzk.mp4 | video_with_audio | 82.066667 | The audio opens with a calm and ambient electronic track, featuring a gentle, rhythmic beat and soft, atmospheric synthesizer melodies that create a serene and modern mood, as the video begins with a wide shot of a luxurious, open-plan living space. The room is filled with natural light, showcasing a harmonious blend o... | 0.75 | 1 | 15 |
finevideo_videos/m3_LboDqetI.mp4 | video_with_audio | 63.096367 | A first-person, wide-angle shot captures a nighttime city scene, initially focused on a green and white sign on a concrete pillar with a brightly lit storefront in the background, all set against the ambient sounds of urban traffic and faint chatter. The camera pans left to reveal a man in a dark jacket standing on a s... | 0.62 | 1 | 10 |
finevideo_videos/ltJslBpy6IA.mp4 | video_with_audio | 170.336833 | The video opens with a static shot of a title card on a white background, featuring a stylized red "R" logo for "RAJANI MD" with a red prescription "Rx" symbol, while the text "ViPeel for Acne" is centered between two horizontal lines. Simultaneously, an upbeat, modern, and slightly electronic instrumental track begins... | 0.88 | 1 | 17 |
finevideo_videos/FHHN0IsuF04.mp4 | video_with_audio | 58.314616 | A static, close-up shot shows a colorful, cartoon-style diagram of an animal cell against a plain white background, outlined in orange with various organelles inside, including a large purple nucleus, blue endoplasmic reticulum, orange Golgi apparatus, and green and red mitochondria. A narrator with a calm, clear male ... | 1 | 1 | 9 |
finevideo_videos/pSgIa8NPFQ4.mp4 | video_with_audio | 163.433333 | A close-up, static shot shows an EpiPen auto-injector against a plain white background, its label clearly visible with the brand name "EpiPEN," "0.3mg EPINEPHRINE AUTOINJECTOR," and "For Allergic Emergencies (Anaphylaxis)." As a red banner with a white thumbs-up icon and "Subscribe" text appears at the bottom, a female... | 1 | 1 | 21 |
finevideo_videos/mkWEZQpl_KA.mp4 | video_with_audio | 108.441667 | A male news reporter with short brown hair, wearing a dark suit and a patterned tie, stands outdoors in a residential neighborhood, speaking directly to the camera as he is surrounded by a crowd. Simultaneously, a news banner at the bottom of the screen reads, "SNOWBALL SHOOTING" and "Feltonville," with a "LIVE" graphi... | 0.75 | 0.89 | 18 |
finevideo_videos/W0zjyqRSINY.mp4 | video_with_audio | 44.76 | In a close-up shot, a man with blondish hair and a goatee, wearing a black and white striped shirt, speaks with an animated and passionate expression against a background of a light-colored wall and a stone-textured surface. As he speaks, a yellow banner at the top reads, "JOSH HOMME FROM QOTSA LOVES BRITNEY SPEARS' TO... | 1 | 0.9 | 10 |
finevideo_videos/x6P8zaJ2TnA.mp4 | video_with_audio | 58.558559 | From a first-person perspective in the video game *Splatoon 3*, the player navigates a narrow corridor with light blue walls splattered in orange and blue ink, with the game's timer and team scores visible at the top. Simultaneously, an off-screen narrator with an energetic and enthusiastic voice calls out, "Hey. Hey y... | 0.75 | 0.83 | 12 |
finevideo_videos/zJUV2ZJP-Ac.mp4 | video_with_audio | 58.641917 | In a medium close-up, a woman with dark hair pulled back, wearing a dark blue t-shirt that reads "WE CLEAN UP LIFE'S ACCIDENTS," stands in front of a large blue truck with "SPAULDING DECON" in white letters and begins to speak with a calm, professional tone. As she says, "So here we are. This is our uh second hoard of ... | 1 | 0.92 | 12 |
finevideo_videos/7H9RMHRER7g.mp4 | video_with_audio | 42.48 | A view from inside a vehicle looks out onto a paved area with yellow lines and a coiled black hose on the ground as a male narrator, who is a firefighter in a bright orange uniform and a white helmet with a red "CFA" logo, gets out of the vehicle and gestures while speaking with a serious and earnest tone, "I've just c... | 1 | 1 | 11 |
finevideo_videos/QKu8ui8Wva4.mp4 | video_with_audio | 37.733333 | A man with dark hair and glasses, wearing an orange shirt, appears in a small circular window in the bottom-left corner of the screen as the main part of the screen displays a web application interface for trip planning. The page is titled "UPDATE TRIP" for a trip called "DEMO - JORDAN FOR THE ADVENTUROUS." With the "P... | 0.88 | 0.89 | 9 |
finevideo_videos/nW4GwmAvfvs.mp4 | video_with_audio | 23.8238 | In a medium close-up shot from a static camera, a woman with blonde, shoulder-length hair, wearing a tan jacket over a black shirt and dark, circular earrings with a green hue, is seated in a studio. Behind her, a large screen displays a view of the Earth from space, showing its blue oceans and white clouds against the... | 0.62 | 1 | 4 |
finevideo_videos/oVwcTll6Yy8.mp4 | video_with_audio | 58.966667 | In a close-up shot, a woman with dark, shoulder-length hair and a red top, identified by an on-screen graphic as "Dr. Caroline Zickgraf, Social Scientist," speaks with an earnest expression while gesturing with her hands. Against a softly lit, out-of-focus interior, she says with a clear, emphatic voice, "Climate chang... | 1 | 1 | 10 |
finevideo_videos/RGS_8DkFREg.mp4 | video_with_audio | 49.333333 | As a narrator with a direct and slightly dramatic tone says, "We're here in South Tampa at a level 5 hoard," a close-up, shaky shot pans across a room filled with an overwhelming amount of clutter, including cardboard boxes with handwritten labels like "Keep Sake Books," a blue plastic bag, and a box of "Lay's" potato ... | 0.75 | 0.9 | 20 |
YAML Metadata Warning:The task_categories "video-to-text" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other
EVA-data
EVA-data is an audiovisual video-caption annotation dataset introduced in EVA-Cap: Optimizing Audiovisual Video Captioning via Event-Centric Alignment. It supports supervised fine-tuning (SFT) and event-centric reinforcement learning for detailed audiovisual captioning.
This release contains 62,881 SFT annotations and two stage-specific GRPO files with 3,200 annotations each. It distributes generated captions, video-path references, quality-control metadata, and, for GRPO, structured event and synchronization-query annotations. Video and audio files are not included.
Video Sources
The videos originate from LLaVA-Video, FineVideo, and Shot2Story. Download the corresponding videos separately from the linked sources and preserve the filenames referenced by video_path.
| Source | Download | Local video directory | SFT annotations |
|---|---|---|---|
| LLaVA-Video | LLaVA-Video-178K | llava_videos/ |
28,425 |
| FineVideo | finevideo-256 | finevideo_videos/ |
14,135 |
| Shot2Story | shot2story-videos | shot2story_videos/ |
20,321 |
| Total | 62,881 |
The source names describe the provenance of the videos; the caption annotations in this release are generated by the EVA-data annotation pipeline rather than copies of the source datasets' original captions.
Dataset Construction
Section 3.1 of the paper describes collecting approximately 100K audiovisual videos, each shorter than 200 seconds, and processing them through three stages:
- Initialization: AVoCaDO generates initial audiovisual descriptions. Samples with repetitive descriptions or abnormal text lengths are filtered out.
- Refinement: Gemini-2.5-Pro receives both the video and its initial description to improve audiovisual temporal alignment and event attribution while preserving the narrative style.
- Event-centric quality control: Gemini-2.5-Pro extracts timestamped audiovisual events from the raw video. GPT-4.1 verifies whether the refined caption covers those events with consistent attribution and synchronization. Samples with an aligned event hit ratio below 0.8 are discarded.
The paper summarizes the resulting dataset as 63K video-caption pairs, with a mean video duration of 73.7 seconds and mean caption length of 504.3 words. These are paper-reported statistics; the exact annotation counts for the files in this release are listed below. The paper also analyzes audio and visual event distributions on a random 20K-video sample and reports diverse video categories, including entertainment, lifestyle, science and technology, sports, and education.
GRPO Reference Annotations
Sections 3.2 and A.2 of the paper describe constructing event-centric reference structures for rollout and GRPO training. Qwen3.5-27B extracts audio and visual events from reference captions and constructs audio-to-visual synchronization queries, including nearby visual events as challenging candidates. Rollout-based selection produces approximately 3.2K samples for each GRPO stage.
The two released GRPO files support different reward combinations:
- Stage 1: Global Event Completeness and Inter-Event Synchronization.
- Stage 2: Global Event Completeness, Intra-Event Attribution, and Inter-Event Synchronization.
GRPO annotations are selected from the SFT video pool. They are training-stage selections, not independent validation or test splits.
Released Files and Counts
Each file is UTF-8 JSONL, with one JSON object per non-empty line. Counts below are measured from the released files, not rounded paper figures. Each file has one annotation per unique video_path.
| File | Training use | Records | Video-source distribution |
|---|---|---|---|
llava_caption_sft.jsonl |
SFT | 28,425 | LLaVA-Video: 28,425 |
finevideo_caption_sft.jsonl |
SFT | 14,135 | FineVideo: 14,135 |
shot2story_caption_sft.jsonl |
SFT | 20,321 | Shot2Story: 20,321 |
stage1_caption_rl.jsonl |
GRPO stage 1 | 3,200 | FineVideo: 3,200 |
stage2_caption_rl.jsonl |
GRPO stage 2 | 3,200 | LLaVA-Video: 1,763; FineVideo: 1,301; Shot2Story: 136 |
The three SFT files have no overlapping video paths. Both GRPO files are subsets of the combined SFT pool; 504 video paths are shared between the two GRPO stages. Do not sum all five files to estimate the number of distinct videos.
Annotation Format
Common Fields
| Field | Type | Description |
|---|---|---|
video_path |
string | Video path relative to the configured video root, such as llava_videos/<original_filename>.mp4. |
caption |
string | Detailed generated audiovisual description used as the reference caption. |
data_type |
string | video_with_audio in this release. |
duration |
number | Video duration in seconds. |
event_total |
integer | Event-count metadata retained from the annotation pipeline. |
event_check_ratio |
number | Event-verification metadata retained from the annotation pipeline. |
qa_check_ratio |
number | QA-check metadata retained from the annotation pipeline. |
The training loader reads video_path and caption; the quality-control fields are retained as metadata rather than dynamically recomputed during training.
A minimal illustrative SFT record is:
{
"video_path": "llava_videos/example.mp4",
"data_type": "video_with_audio",
"caption": "A person explains how to use a tool while demonstrating the steps."
}
This example illustrates the format only; example.mp4 is not a provided video. Released records also include the duration and quality-control fields listed above.
Additional GRPO Fields
| Field | Type | Description |
|---|---|---|
event_list |
object | Contains audio and visual event lists. Each event includes modal, type, character, and content. Some stage-2 audio events also retain character_former. |
a2v_query_list |
array | Audio-to-visual synchronization queries used for the alignment reward. |
Each synchronization query contains the following fields:
| Field | Description |
|---|---|
modal, type, character, content |
The audio event serving as the query. |
visual_candidates_for_judge |
Five visual-event objects, each with modal, type, character, and content. |
selected_visual_indices_new |
Reference visual indices consumed by the current alignment-reward implementation; candidate positions in the judge prompt are numbered from 1. |
selected_visual_indices, window_range |
Additional selection metadata retained from reference construction. |
The released stage-1 records contain 3-30 synchronization queries per video; stage-2 records contain 3-25. Preserve the supplied candidate order and reference indices when using these files. The release does not contain a sync_event_list field; the current loader treats that field as optional.
Data Preparation for EVA-Cap
Download the five JSONL files into the project-root data/ directory of the EVA-Cap code repository. Download the source videos separately and arrange them as follows:
EVA-Cap/
|-- data/
| |-- llava_caption_sft.jsonl
| |-- finevideo_caption_sft.jsonl
| |-- shot2story_caption_sft.jsonl
| |-- stage1_caption_rl.jsonl
| |-- stage2_caption_rl.jsonl
| |-- llava_videos/
| |-- finevideo_videos/
| `-- shot2story_videos/
`-- data_config/
|-- sft_caption_datasets.json
|-- rl_caption_datasets_stage1.json
`-- rl_caption_datasets_stage2.json
Keep original filenames and any subdirectories required by video_path. The default training configurations set video_root to data, so llava_videos/example.mp4 resolves to data/llava_videos/example.mp4 when running from the project root.
| Training stage | Code-repository configuration | Annotation files |
|---|---|---|
| SFT | data_config/sft_caption_datasets.json |
The three *_caption_sft.jsonl files |
| GRPO stage 1 | data_config/rl_caption_datasets_stage1.json |
stage1_caption_rl.jsonl |
| GRPO stage 2 | data_config/rl_caption_datasets_stage2.json |
stage2_caption_rl.jsonl |
If your videos or annotations are stored elsewhere, update video_root and annotation in these configuration files. For GRPO, use the supplied stage-specific annotations, including event_list and a2v_query_list; plain SFT caption records do not provide the required event-centric reward references.
See the EVA-Cap training instructions for the processor patch, checkpoint selection, reward-server setup, and training commands.
Limitations and Source Terms
Captions and reference structures are automatically generated and verified. They may still contain omissions, hallucinations, incorrect speaker attribution, or inaccurate temporal relationships. They are not guaranteed human-verified ground truth.
Source videos remain subject to their original licenses, access conditions, and any underlying video-owner rights. Follow the terms of each linked source when downloading, using, or redistributing videos or derived content. This annotation release does not grant rights to redistribute the source videos. The EVA-Cap code license does not automatically license the data; consult the dataset release's applicable annotation-license declaration separately.
References
- EVA-Cap: Optimizing Audiovisual Video Captioning via Event-Centric Alignment: Sections 3.1, 3.2, and Appendix A.2 describe dataset construction and event-centric training.
- LLaVA-Video: Video Instruction Tuning with Synthetic Data.
- FineVideo: The source audiovisual video dataset used for a portion of EVA-data.
- Shot2Story: A New Benchmark for Comprehensive Understanding of Multi-Shot Videos.
- AVoCaDO: An Audiovisual Video Captioner Driven by Temporal Orchestration, used for initial caption generation.
- Downloads last month
- 35