encoder-decoder-based-video-captioning's Introduction

Encoder-Decoder-based Video Captioning

This repository provides an Encoder-Decoder model to generate captions for input videos.

The ability to be applied for numerous applications mark Video Captioning's importance. For example, it can be applied to help search videos across web pages in an efficient manner and it can also cluster the videos having a large degree of similarity in terms of their respective generated captions.

Requirements

Tensorflow
Keras
OpenCV
NumPy
FuncTools

Usage

Data

The MSVD dataset developed by Microsoft can be downloaded from here.
This data set contains 1450 short YouTube clips that have been manually labeled for training and 100 videos for testing.
Each video has been assigned a unique ID and each ID has about 15–20 captions.

Training and Testing

To extract features for frames of every single input videos using pre-Trained VGG model, run Extract_Features_Using_VGG.py.
To train the developed model, run training_model.py.
To use the trained Video Captioning model for inference, run predict_model.py.
To use the trained model for real-time Video-Caption generation, run Video_Captioning.py.

Results

Following are a few results of the developed Video Captioning approach on test videos:-

Test Video	Generated Caption
	a woman is mixing some food
	a man is performing on a stage
	a man is mixing ingredients in a bowl
	a man is spreading a tortilla
	a woman is seasoning some food
	a cat is playing the piano

Recommend Projects