INSTALL.md

Installation and Preparation

Requirements

SceneSeg is currently very easy to install before the introduction of feature extractor

• Python 3.6+
• PyTorch 1.0 or higher
• mmcv

a. Install PyTorch and torchvision following the official instructions, e.g.,

conda install pytorch torchvision -c pytorch

b. Clone the SceneSeg repository.

git clone https://github.com/AnyiRao/SceneSeg.git
cd SceneSeg

c. Install Python Packages

pip install -r docs/requirements.txt

Folder Structure

|-data ## the data_root for experiments
|-run  ## to store experiments
|-pre  ## for preprocess
|-lgss
|   |-src
|   |-config
|   |-utilis
|   |-run.py
|   |-gen_csv.py

Download Pretrained Models

Pretrained models are at Google Drive. Please follow the provided folder structure and put it under run.

Prepare Datasets for Scene318

Annotation and metadata are at Google Drive. The full movienet annotation including, debut year, cast, director info and so on can be downloaded from OpenDataLab

The intermediate features place_feat, cast_feat, act_feat, aud_feat are located at OneDrive due to the limited free size of Google Drive. They can be extracted with unzip e.g., unzip act_feat.zip.

Put label318 shot_movie318 meta/split318.json, intermediate features place_feat, cast_feat, act_feat, aud_feat under data.

Explanation

label318 is the scene transit (1) or not (0) label for each shot. shot_movie318 is the shot and frame correspondence to recover the time of each scene. They will be automatically handled by the processing codes e.g., the data_pre function in src/data/all.py

Prepare Your Own Dataset

If you run our demo, you don't need to prepare. Just run our three lines.

If you wish to use the method's full functions, you may need to organize your data like ours.