Frame Selection Strategy

Description

This project is for selecting frame of all kinds of videos based on multiple frame scoring policy. Aims at accerlating the pre-processing step of Computer Vision works, we integrated different frame selection policies and developed this flexible and fast frame selection lab.

Purpose

• Developed a useful tool for extracting frame from videos in a flexible way.
• An extensible architecture following Registry Mode for further development.
• Visuable exhibition of the selection result.

Project Architecture

• registry: contain source code for registries and their configs
• score
• strategy

Development Plan

• Develop evaluation methods for each strategy
• Develop API and API document
• Develop OOD-based scores and strategy

Using Guidance

Quick Run

1. Clone the specific branchs with

# bash
git clone https://github.com/JasonZuu/Frame-Selection.git

1. import the abstact class.

# python
from frame_selection import FrameSelectionAbstract

3. Call functions of FrameSelectionAbstract to do whatever you want.

# python
fs_abstract = FrameSelectionAbstract()

# check the scorers and critor you can work with
keys_infos = fs_abstract.reset(help_mode=True)
print(keys_infos)

# reset the FS
fs_abstract.reset(video_path="my_unittest/test.mp4")

# score frames
scores = fs_abstract.score(group_size=24, resize_shape=(64,64))
print(scores[:10])

# select frames
selected_frames, selection_score = fs_abstract.select(select_num=10)
print(selection_score)

# export frames
frame_paths = fs_abstract.export(export_dir="data")
print(frame_paths[:5])

API

All functions of the FS has been wrapper by abstract.FrameSelectionAbstract .

Document

Basic Descrip of modules