Author(s: Dakshtrehan Computer Vision, Machine Learning How DeepFakes can be created without any programming. Are you a fan of artists and performers? Are you interested in imitating them? You can now! Are you a TikTok user but afraid to show off your dancing moves? AI has your back. You just need a still photo and a clip of your favourite artist doing some moves. And, BOOM! You’re now an Internet sensation. Image animation is a method that uses neural networks to imitate this workflow. It wraps the source image around the object’s motion in the video sequence. This article will show you how simple image animation can be and what code is required to use it. It works in this way: This model, “First Order Motion Model for Image Animation”, is an innovative approach to replacing the object in driving video with the original image. It does not require any further information. It is important that you understand the workings of your sequence before starting to construct it. A neural network is used to reconstruct the driving video. The object in the video replaces the one in the source picture. The model attempts to predict the movement of the object in the source picture based on the frame representations in the driving videos. The model is able to track every detail of driving video, from head movements to speaking. WorkFlow Training is performed on large amounts of video. The model uses frame pairs to reconstruct the training videos. It then attempts to understand the latent motion of the video by extracting the information. Using the extracted information, it can encode the motion using a combination of native affine transforms and motion-specific key points displacements. The model then recreates the sequence of video by attaching the object from the source image to every frame of driving footage and performing image animation. This framework was implemented with a motion estimation and an image generation module. Motion estimator’s goal is to discover a latent representation for motion within the video. It tracks the motion in the video and encodes them as local affine transforms and key points displacements. It produces a dense motion area and an occlusion map. This mask defines the relevant areas of the driving sequence object that will be covered by the source images (e.g. Front side of the head. In the above gif the back is not animated. This can be attributed to a motion estimator. The output from the motion estimator, along with driving video and source image are then input to the generator. An image generator creates driving video frames with a wrapped source object. The frames can then be joined to form a video sequence. You can easily find the source code for this paper on Github. Two methods can be used to make a deepfake. Old-School Method: You can download the repository as a zip file. 2. Save the files. 3. You can use the terminal to extract the folder, and then download the packages by using the following command. 4. Get a trained checkpoint. You can download any type of checkpoint you like. 5. To the extracted folder, copy source image, driver’s video and pre-trained checkpoint. 6. Open the terminal and use the following command: Replace “config/dataset_name.yml”, “path/to/driving”, “path/to/source”, “path/to/checkpoint” to dataset name, driving sequence path, source image path and checkpoint path respectively. If you do not have a GPU, run the modified command. 7. Your DeepFake video will be in the same folder. Google-Colab Technique: Visit https://colab.research.google.com/github/AwaleSajil/DeepFake_1/blob/master/first_order_model_demo(Youtube)_new_audioV5_a.ipynb Create a copy of this ipynb file. Create your model. 4. Training data is used to train your model. Above command generates output 5. Include your source, and you will have a driving video. 6. Enjoy your DeepFake. References: [1] First Order Motion Model for Image Animation (nips.cc) [2] AliaksandrSiarohin/first-order-model: (github.com) [3] First order model (aliaksandrsiarohin.github.io) [4] How to Produce a DeepFake Video in 5 Minutes | by Dimitris Poulopoulos | Towards Data Science [5] From Zero to Deepfake. Exploring deepfakes with DeepFaceLab | by Jarrod Overson | Medium [6] (51) How To Create Deepfake Videos In 5 Minutes (Hindi) — YouTube Feel free to connect: Portfolio ~ https://www.dakshtrehan.com LinkedIn ~ https://www.linkedin.com/in/dakshtrehan Follow for further Machine Learning/ Deep Learning blogs. Medium ~ https://medium.com/@dakshtrehan Want to learn more? Are You Ready to Worship AI Gods?Detecting COVID-19 Using Deep LearningThe Inescapable AI Algorithm: TikTokGPT-3 Explained to a 5-year old.Tinder+AI: A perfect Matchmaking?An insider’s guide to Cartoonization using Machine LearningReinforcing the Science Behind Reinforcement LearningDecoding science behind Generative Adversarial NetworksUnderstanding LSTM’s and GRU’sRecurrent Neural Network for DummiesConvolution Neural Network for Dummies Cheers Give me 5 minutes, I’ll give you a DeepFake! This story was first published on Medium by . People are responding and highlighting it. Published via
If you give me five minutes I will make you a DeepFake.
