A Python project that uses binary image classification to detect whether individual frames from clips are real or AI generated. Makes use of a neural network which is trained on two classes of data: frames generated using viral Veo3 videos online (Google's latest AI video generational model) alongside frames from real world movie scenes for comparison.
• User inputs an image
• If the file path the user has entered is valid, the image will be passed to the classification model
• The classification model utilises the dataset it was trained on to detect whether a given image represents a frame from a clip that is either real or AI generated
• If the model is uncertain that a given image represents a frame from a clip/movie scene (e.g. probability it belongs to either of the classes is less than 0.9), it is unlikely to be a valid input so we raise an error
• Otherwise, the model will predict whether the frame is real or AI generated
• Scrape data from a wider range of video sharing platforms in order to develop a more accurate model
• Install a GPU for efficiency in testing
• Potentially create more classes to increase model accuracy (e.g. current model may falsely predict that a fictional movie is AI generated)
The model has already been trained, so all you need to do is run this command:
python moviescenedetector.py
Make sure to load the images you want to test into the repository provided, so the program is able to identify the file path.