A Short Note about Kinetics-600

By João Carreira, Eric Noland, Andras Banki-Horvath, Chloe Hillier, Andrew Zisserman. 2018

636 citations recorded by the source corpus.

We describe an extension of the DeepMind Kinetics human action dataset from 400 classes, each with at least 400 video clips, to 600 classes, each with at least 600 video clips. In order to scale up the dataset we changed the data collection process so it uses multiple queries per class, with some of them in a language other than english -- portuguese. This paper details the changes between the two versions of the dataset and includes a comprehensive set of statistics of the new version as well as baseline results using the I3D neural network architecture. The paper is a companion to the release of the ground truth labels for the public test set.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

Citations in the corpus

Scaling Autoregressive Video Models (2019, methodology): “Speciﬁ-cally, we use Kinetics-600, which contains roughly 400K training videos ranging over 600 action classes (Carreira et al., 2018).”
Learning Video Representations from Textual Web Supervision (2020, background): “Kinetics-400, 600, and 700 are increasingly large versions of the dataset, containing 400, 600, and 700 action categories, respectively [7, 8].”
Self-Supervised MultiModal Versatile Networks (2020, methodology): “Our method is simple yet it matches or exceeds the state-of-the-art for action and audio classification on five challenging benchmarks: HMDB51, UCF101, Kinetics600, ESC-50 and AudioSet.”
A Short Note on the Kinetics-700 Human Action Dataset (2019, methodology): “The new dataset follows the same principles as Kinetics-400 [7] and Kinetics-600 [2]: (i) The clips are from YouTube videos, last 10s, and have a variable resolution and frame rate; (ii) for an action class, all clips are from different YouTube videos.”
MERLOT RESERVE: Neural Script Knowledge through Vision and Language and Sound (2022, methodology): “[19] Joao Carreira, Eric Noland, Andras Banki-Horvath, Chloe Hillier, and Andrew Zisserman.”
A Short Note on the Kinetics-700-2020 Human Action Dataset (2020, methodology): “Three editions have been released: Kinetics-400 [6], Kinetics-600 [1] and Kinetics-700 [2], with 400, 600 and 700 human action classes, respectively.”
Scalable Adaptive Computation for Iterative Generation (2022, methodology): “For video prediction, we use Kinetics-600 dataset (Carreira et al., 2018) at 16×64×64 resolution.”
SlowFast Networks for Video Recognition (2018, methodology): “Kinetics-600 [3] has ∼392k training videos and 30k validation videos in 600 classes.”
VATT: Transformers for Multimodal Self-Supervised Learning from Raw Video, Audio and Text (2021, methodology): “Furthermore, we set new records on Kinetics-400 [15], Kinetics-600 [14], Moments in Time [67], and AudioSet [38] without supervised pretraining.”
Is Space-Time Attention All You Need for Video Understanding? (2021, methodology): “In Table 6, we also present our results on Kinetics-600.”
Multiscale Vision Transformers (2021, methodology): “Our focus in this paper is video recognition, and we design and evaluate MViT for video tasks (Kinetics [59, 10], Charades [86], SSv2 [38] and AVA [39]).”
CoCa: Contrastive Captioners are Image-Text Foundation Models (2022, methodology): “Our visual recognition experiments are conducted on ImageNet [9] as image recognition benchmark, and multiple video datasets including Kinetics-400 [57], Kinetics-600 [58], Kinetics-700 [59], Moments-in-Time [60] as test-beds for video action recognition; it is noteworthy that CoCa pretrains on imag”
ViViT: A Video Vision Transformer (2021, background): “Kinetics [1, 4] is a dynamic dataset – it consists of YouTube videos which are specified by their URLs, and it is possible that these videos are removed by their original creators.”
Video Diffusion Models (2022, methodology): “We additionally evaluate video prediction performance on the Kinetics-600 benchmark [27, 9].”
MoViNets: Mobile Video Networks for Efficient Video Recognition (2021, methodology): “We report results on all Kinetics datasets, including Kinetics 400 [9, 32], Kinetics 600 [6], and Kinetics 700 [7], which contain 10-second, 250-frame video sequences at 25 fps labeled with 400, 600, and 700 action classes, respectively.”
Spatiotemporal Contrastive Video Representation Learning (2020, methodology): “We mainly conduct experiments on the Kinetics-400 (K400) [38] and Kinetics-600 [8] (K600) datasets.”
Masked Feature Prediction for Self-Supervised Visual Pre-Training (2021, positive): “Finally, we ﬁne-tune these MViT-L, 16 × 4 models at a larger input resolution of 312 and a longer duration of 40 × 3 to achieve 88.3 % top-1 on K600 and”
SwinBERT: End-to-End Transformers with Sparse Attention for Video Captioning (2021, neutral): “The VidSwin is initialized with Kinetics-600 pre-trained weights [34], and the multimodal transformer encoder is randomly initialized.”
X3D: Expanding Architectures for Efficient Video Recognition (2020, result): “A.1 show the identical results on the test set of Kinetics-400 (which has been publicly released with Kinetics-600 [2]).”
V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning (2025, neutral): “…v2 dataset (SSv2) introduced in Goyal et al. (2017), exo-centric action videos from the Kinetics 400, 600, and 700 datasets (Kay et al., 2017; Carreira et al., 2018, 2019), YouTube tutorial videos from HowTo100M (Miech et al., 2019), and general YouTube videos from YT-Temporal-1B (Zellers et…”