There are 2 versions of MGif dataset: raw and processed.
The raw dataset is composed of the .gif files. Files is manually downloaded using google search, we use the following search phrase "cat cycle gif animation". And replace the cat with different animals such as dog, elephant, camel and so on. The original names is preserved. The dataset can be downloaded.
In processed dataset bg is manually removed using the tool bg_removal_tool.py.
Usage:
python bg_removal_tool.py /folder/with/raw/gifs /folder/for/processed/gifs /folder/for/bad/gifs/which/you/want/to/exclude
All the gifs is resized to 256 x 256. The files is renamed, correspondence between raw and processed could be established via mapping.txt file. The dataset can be downloaded.