The Hopkins 155 dataset was introduced in [1] and has been created with the goal of providing an extensive benchmark for testing feature based motion segmentation algorithms. It contains video sequences along with the features extracted and tracked in all the frames. The ground-truth segmentation is also provided for comparison purposes. For a more comprehensive description of the dataset, please refer to the main Hopkins 155 page.

Each sequence in the Hopkins 155 dataset contains only complete trajectories and no outliers. We provide 16 additional sequences (used in [2], [3] and [4]) that present missing data and outliers. These sequences have the same format as the Hopkins 155 sequences. We refer to the entire dataset (standart plus the additional sequences) as the Hopkins 155+16 dataset. The additional sequences have been made publicly available on April 20th, 2010.

Description page for the Hopkins 155+16 dataset

Results of some motion segmentation algorithms on the Hopkins 155 dataset in our research

Download Hopkins 155

Dataset without videos
Dataset with videos (Part 1)
Dataset with videos (Part 2)
Dataset with videos (Part 3)
Dataset with videos (Part 4)
Dataset with videos (Part 5)
Dataset with videos (Part 6)
Dataset with videos (Part 7)
Dataset with videos (Part 8)

Download Hopkins 155+16 Additional Sequences

Additional Sequences with Missing Data
Additional Sequences with Missing Data and Outliers (Part 1)
Additional Sequences with Missing Data and Outliers (Part 2)
Additional Sequences with Missing Data and Outliers (Part 3)

List of Sequences of Each Category

In order to test the robustness to outliers of segmentation algorithms, one can add synthetically generated outliers to the trajectories in each sequence of the Hopkins 155 dataset. We provide MATLAB code that generates these outlying trajectories for any sequence.

Code for generating synthetic outliers

This package is a subset of the Hopkins 155 dataset and contains only the sequences used in [6].

Download Three-view Segmentation Sequences

We are making available the "Hands" sequence, a dataset (first appeared in [5]) used for testing Non-Rigid Structure from Motion algorithms.

Download Hands sequence

Existing dynamic texture databases are not well-suited for testing joint segmentation and categorization algorithm. This is because most of the video sequences in these databases either contain only a single texture and seldom any background. We have annotated 117 videos at the pixel level from the 3 largest classes of the Dyntex database: waves, flags and fountains. This mat file contains the pixel wise annotations.

Download Annotated Dynamic Texture Segmentation Dataset

For more details please refer to Joint Segmentation and Categorization of Dynamic Textures