Skip to content

Dataset · demalenk

Point tracking video benchmark

demalenk/itto-dataset

An image dataset from three video sources for evaluating and diagnosing point tracking methods.

What's inside

The ITTO benchmark includes between 1,000 and 10,000 samples combining data from three sources: MOSE, L-VOS, and Ego4D. The materials are presented in image modality and designed to test tracking algorithms. Due to licensing restrictions, the L-VOS and Ego4D portions must be installed separately.

What you can do with it

  • Evaluating the quality of point tracking models on dynamic scenes
  • Diagnosing weaknesses of computer vision algorithms during object tracking
Size
1K–10K rows
Downloads this month
30.1K
Likes
5

How to load it

from datasets import load_dataset

ds = load_dataset("demalenk/itto-dataset")