The PortraitMode-400 dataset is a significant contribution to the field of video recognition, specifically focusing on portrait mode videos. Let me provide you with more details:

1. Dataset Overview:
2. The PortraitMode-400 (PM-400) dataset is the first of its kind and is dedicated to portrait mode video recognition.
3. It was created to address the unique challenges associated with recognizing videos captured in portrait mode.

4. Portrait mode videos are increasingly important due to the growing popularity of smartphones and social media applications.

5. Data Collection and Annotation:

6. The dataset consists of 76,000 videos collected from Douyin, a popular short-video application.
7. These videos were meticulously annotated with 400 fine-grained categories.

8. Rigorous quality assurance measures were implemented to ensure the accuracy of human annotations.

9. Research Insights and Impact:

10. The creators of the dataset conducted a comprehensive analysis to understand the impact of video format (portrait mode vs. landscape mode) on recognition accuracy.
11. They also explored spatial bias arising from different video formats.
12. Key aspects of portrait mode video recognition were investigated, including data augmentation, evaluation procedures, the importance of temporal information, and the role of audio modality.

(1) [2312.13746] Video Recognition in Portrait Mode - arXiv.org. https://arxiv.org/abs/2312.13746. (2) Video Recognition in Portrait Mode | Papers With Code. https://paperswithcode.com/paper/video-recognition-in-portrait-mode. (3) Video Recognition in Portrait Mode - arXiv.org. https://arxiv.org/pdf/2312.13746.pdf. (4) undefined. https://doi.org/10.48550/arXiv.2312.13746.