Skip to content

Latest commit

 

History

History
41 lines (34 loc) · 1.59 KB

DATASET.md

File metadata and controls

41 lines (34 loc) · 1.59 KB
Raw

Dataset Preparation

Training Dataset

We train our model on the VLOG Dataset. We use the official release of the videos, the files are named as "block_x.tar" (0<=x<=4). We assume the videos are downloaded on the path: YOUR_DATASET_FOLDER/vlog/.

Download the list for the videos data_v1.1.tgz. Extract the list "manifest.txt" to the same folder: YOUR_DATASET_FOLDER.

Go into the folder:

    cd preprocess

Change the video path in preprocess/downscale_video_joblib.py. Reduce the video size and save it to YOUR_DATASET_FOLDER/vlog_256/ :

    python downscale_video_joblib.py

Extract the jpgs to YOUR_DATASET_FOLDER/vlog_frames_12fps/ by using:

    python extract_jpegs_256.py

Gnerate the jpg list to YOUR_DATASET_FOLDER/vlog_frames_12fps.txt for training:

    python genvloglist.py

Testing Dataset

We test our model on the DAVIS 2017 dataset in this repo. We assume the dataset is downloaded on the path: YOUR_DATASET_FOLDER/davis/ . Clone the evaluation code for DAVIS 2017 to YOUR_DATASET_FOLDER/davis-2017/ .

Go into the folder:

    cd preprocess

Generate the list for testing as YOUR_DATASET_FOLDER/davis/DAVIS/vallist.txt :

    python gendavis_vallist.py