dung-huynh / tags

tags, video

tags: a website that takes your video and annotate it. You just need to give a video link and our algorithm will deliver you a set of words for each scene with given interval (frame start, frame end).

Definition of annotate: provide words that describe an object, this case, a frame, a scene, or the video.

Features:

• annotate scenes: frames that share similarities and settings i.e. environment or context.
• annotate video: (ultimate goal)

Contributors:

• Dung Huynh
• Henok Seifu
• Melvin Wolf
• Osama Al-Wardi
• Saurav Dahal

The project was started during jacobsHack!

tags's Issues

[front-end] website

Features:

• text box for video link and check whether the link is valid (video)
• download button to get the video to a local folder
• text: : <word_1>, <word_2>, ...

[feature] splitting a video into smaller set of frames

Take a frame for every 10 frames

Flag special tags

Such as unwanted tags (violence)

[merging-features] get video from website -> split frames

[feature] create a list of words for each frame

Input: the folder containing all frames
Output: a list of lists of words. Each list of word describes a frame.

[feature] posting json file

[feature] counting words

Output: A descending dictionary where each key-value pair is (word, count) regarding words appearances.

Purpose: to detect outliers i.e. logos, watermarks,...

[front-end] add input to adjust the amount of skip (for frames)

[feature] merge scenes to context (second layer grouping)

[merging] clueing everything

[feature] converting lists into formatted json file

[feature] separating scenes

Scenes are separated by absolute changes.
Frames are separated by relative changes.

Changes need to be distinguished such that the program knows when to create a new group (scene).

[feature] find a library that puts words into sentences

[feature] regrouping

After a black list of words is given, the groups (scenes) are redistributed.

Create sentences from words to get a true description

[feature] checking for outliers

Find the word(s) that appear all the time or many times (?) across all groups. If there is such a word, this word should be eliminated since it doesn't contribute into a scene.

Be careful that some might appear a lot in scenes (maybe, the word: "people"), hence, if the word appears all the time and outnumber the rest.

[feature] determine whether two words (fed as inputs of a function) are relatable