Misinformation-Video-Detection

Final project for the Computer Vision course 2024

Step 1:

Downloading and downsampling the data.

Step 2:

Preprocessing data: converting the annotation file URLs into an image dataset constisting of youtube video thumbnails.

Step 3:

Image resize: Resizing all the thumbnail images to a unifrom 320x320.

Step 4:

Dataset cleaning: filtered only relevant attributes from the csv file. Reduced the thumbnail dataset samples to match the filtered annotations.

Step 5:

Linear Classifier: implemented LogisticRegression model using sklearn.

Step 6:

Nerual Network: implemented basic and complex NN model using Tensorflow.

Step 7:

CNN: implemented CNN models with and without regularisation. Also implemented pretrained VGG16 and resnet models (experimented with regularisation and finetuning).

Step 8:

Increasing the dataset size: the models were run on the larger dataset.

Here is the link to the downsampled dataset:

https://drive.google.com/drive/folders/1LLrRdI3bqBoqQ5Wul8DM9DGFgPbSsEKv?usp=sharing