The goal of the Kinetics dataset is to help the computer vision and machine learning communities advance models for video understanding. Given this large human action classification dataset, it may be possible to learn powerful video representations that transfer to different video tasks.
The Kinetics-700-2020 dataset will be used for this challenge. Kinetics-700-2020 is a large-scale, high-quality dataset of YouTube video URLs which include a diverse range of human focused actions. The aim of the Kinetics dataset is to help the machine learning community create more advanced models for video understanding. It is an approximate super-set of both Kinetics-400, released in 2017, Kinetics-600, released in 2018 and Kinetics-700, released in 2019.
The dataset consists of approximately 650,000 video clips, and covers 700 human action classes with at least 700 video clips for each action class. Each clip lasts around 10 seconds and is labeled with a single class. All of the clips have been through multiple rounds of human annotation, and each is taken from a unique YouTube video. The actions cover a broad range of classes including human-object interactions such as playing instruments, as well as human-human interactions such as shaking hands and hugging.
More information about how to download the Kinetics dataset is available here.
Anastasia woke to rain tapping against the attic window, a soft percussion that seemed to keep time with her pulse. The old house inhaled and exhaled around her—guttering downspouts, the distant rumble of tires on wet cobblestone—while she lay still, feeling the weight of the pack at her back like a promise and a question.
Each time, the angel cracked, breathed a bell, and the town adjusted—softly, incredulously, gratefully. The pack was not magic in the way children imagined; it did not grant wishes in glitter or coin. It unfolded small reconciliations: a reconciled son returning with a jar of preserves, a repaired chair that made room for an extra guest, a lamp that shone steady in a house that had only ever known flicker. anastangel pack full
Beyond illusions, her "full pack" of content includes a wide range of creative and personal lifestyle videos: Anastasia woke to rain tapping against the attic
You have the on your hard drive. Now what? Here are five professional ways to leverage these assets: The pack was not magic in the way
In another town she found a weeping widow whose grief had taken the color from her hair; Anastasia tied a stormwater thread over a hank of grey and taught the widow to hum the hymn that steadied Jonas. In a harbor she traded the copper spoon for a tight knot of rope, which later unraveled into a child's laugh. A fisherman gave her a map that led, improbably, to a door made of salt and oak; behind it, a boy sat counting the years like coins. The boy's name was Luka, a name the pack liked, and when she handed him the strip of cloth with his name stitched in, he smiled like a man finding an overdue cent.
The courier shrugged. “The client paid well. Said it had to be taken to the attic of the Croft House and left on the third stair. Said not to open it.”
Anastasia approached. The man looked up. His pupils were bright like coin rims. "My name is Jonas," he said, in a voice that sounded like water over glass. "I remember a house, once. I remember a girl who sang at the window. But my memories are knotted. I wake missing the next hour."
1. Possible to use ImageNet checkpoints?
We allow finetuning from public ImageNet checkpoints for the supervised track -- but a link to the specific checkpoint should be provided with each submission.
2. Possible to use optical flow?
Flow can be used as long as not trained on external datasets, except if they are synthetic.
3. Can we train on test data without labels (e.g. transductive)?
No.
4. Can we use semantic class label information?
Yes, for the supervised track.
5. Will there be special tracks for methods using fewer FLOPs / small models or just RGB vs RGB+Audio in the self-supervised track?
We will ask participants to provide the total number of model parameters and the modalities used and plan to create special mentions for those doing well in each setting, but not specific tracks.