SportsGrounding Dataset

1. Task Description

The Spatio-Temporal Video Grounding (STVG) task aims to take an untrimmed video and a natural language description as input, and output the spatio-temporal tube corresponding to the natural language description—specifically, locating the start and end frames, as well as the bounding boxes of the target within the located segment.

SportsGrounding is based on the basketball subset of the MultiSports dataset, focusing on basketball-related scenarios. It contains numerous complex interaction scenes between people and between people and objects, with a larger number of instances.

2. Data Overview

The dataset is modified based on the video and annotation data of MultiSports. Six videos with excessive repetitive actions that cannot be distinguished by natural language are removed, resulting in a total of 520 videos. It follows the train/val split of MultiSports, with 374 videos in the training set and 146 videos in the validation set. Unlike other STVG datasets, each video in SportsGrounding contains multiple captions describing different target persons.

Key Statistics of SportsGrounding

Metric	Value
Dataset Size	4243 tubes (i.e., 4243 video-text pairs)
Average Video Duration	19.70 seconds
Average Tube Duration	1.49 seconds
Average Description Length	16.89 words

Comparison with Other Datasets

Metric	VidSTG	HC-STVG v1	HC-STVG v2	SportsGrounding
Data Source	VidOR	AVA	-	MultiSports
Dataset Size	99943 video-text pairs	5660 video-text pairs	16544 video-text pairs	4243 video-text pairs
Average Video Duration	28.01 seconds	20 seconds	-	19.70 seconds
Average Tube Duration	9.68 seconds	5.37 seconds	-	1.49 seconds
Average Description Length	Declarative: 11.12; Interrogative: 8.98	17.25	-	16.89

Unique Characteristics of SportsGrounding

• Some instances have very short durations.
• More people appear in the videos (other datasets contain fewer people even in multi-person scenarios; 57.2% of videos in HC-STVG have more than 3 people, and the rest have 2 people).
• More complex interactions between people; many descriptions require certain reasoning and scene information modeling. Example: "The defender in white is blocked by the teammate of this offensive player."

3. Data Format

Annotation files are divided into train.json and val.json. Each annotation entry is structured as follows:

{
  "v_00HRwkvvjtQ_c001.mp4": [
    {
      "bbox": [[x0, y0, w0, h0], [x1, y1, w1, h1], ...],  # Continuous bounding boxes of the target person
      "fps": 25,  # Frames per second of the video
      "st_frame": 7,  # Ground truth start frame (index starts at 1)
      "ed_frame": 28,  # Ground truth end frame
      "caption": "The player in white attempts to gain control of the ball after an official tosses it into the air between him and his opponent in blue.",  # Caption describing the target person
      "width": 1080,  # Image width
      "height": 720  # Image height
    },
    ...
  ],
  ...
}