Table 4

A summary of common small-scale datasets from 2011 to now used for action recognition.

DatasetDescription#classesSamplesDownload
HMDB51 [59]

- At least 1s / video.

- Single activity / video.

516,849Link
UCF50 [88]

- Realistic videos from Youtube.

- Single activity / video.

506,676Link
UCF101 [98]

- At least 1.06s/video.

- Single activity / video.

10113,320Link
ActivityNet [9]

- Large-scale video.

- 1.41 activity instance / video.

20327,811Link
Hollywood2 [72]- 19.7s/video on average action videos and scene videos.223,669Link
MSR-Action3D [64]An action dataset of depth sequences captured by a depth camera.20Link
MSR-Daily Activity 3D [112]

- A daily activity dataset captured by a Kinect device camera.

- An activity is performed in either “sitting on sofa” or “standing” pose.

12320Link
ASLAN [55]- Focus on action similarity.4323,697Link
RGBD-HuDaAct [80]- Synchronized color-depth video streams 30s-150s/video.161,189Link
Charades [93]- Video action classification performance 6.8 actions/video.1579,848Link

or Create an Account

Close Modal
Close Modal