Action Recognition in the Dark

Download this dataset as a .zip file

Action Recognition in the Dark Dataset (ARID Dataset)


A Benchmark Dataset for Action Recognition in Dark Videos

alt text

Updates (ARID v1.5)

ARID has been honored to be the benchmark dataset of the UG2+ Challenge 2021/2022 held in conjunction with CVPR 2021/2022. Over this period we have updated ARID by expanding ARID to include more scenes and actions to better facilitate action recognition in low illumination, through both supervised and semi-supervised methods. The ARID (v1.5) has been released as shown here. It now contains 5,572 clips with more than 320 clips per action, with a total of 11 actions. These clips are shot in 24 scenes (12 indoor, 12 outdoor) with more than 15 volunteers. The ARID (v1.5) is also part of the Daily-DA dataset. The ARID (v1) dataset is still available for download!

Abstract (ARID v1)

The task of action recognition in dark videos is useful in various scenarios, e.g., night surveillance and self-driving at night. Though progress has been made in action recognition task for videos in normal illumination, few have studied action recognition in the dark. This is partly due to the lack of sufficient datasets for such a task. In this paper, we explored the task of action recognition in dark videos. We bridge the gap of the lack of data for this task by collecting a new dataset: the Action Recognition in the Dark (ARID) dataset. It consists of over 3,780 video clips with 11 action categories. To the best of our knowledge, it is the first dataset focused on human actions in dark videos. To gain further understanding of our ARID dataset, we analyze the ARID dataset in detail and showed its necessity over synthetic dark videos. Additionally, we benchmark the performance of several current action recognition models on our dataset and explored potential methods for increasing their performances. Our results show that current action recognition models and frame enhancement methods may not be effective solutions for the task of action recognition in dark videos.

Basic Statistics

The distribution of clips among the 11 classes is as follows:

alt text

Comparisons with HMDB51(-dark)

We compare our ARID dataset statistically with HMDB51/HMDB51-dark, with the results and sampled frame as shown:

alt text

alt text

Benchmark Results

Here we present some benchmark results of previous action recognition models: (Across three splits)

Method Top-1 Accuracy Top-5 Accuracy
VGG-Two Stream 32.08% 90.76%
TSN 57.96% 94.17%
C3D 40.34% 94.17%
I3D-RGB 54.68% 97.69%
I3D-Two Stream 72.78% 99.39%
3D-ResNet (50) 71.08% 99.39%

The 5th UG2+ Challenge Workshop

The 4th UG2+ Challenge Workshop

Papers and Download

CC BY 4.0

Back to Project Page