Top 10 Machine Learning Demos: Hugging Face Spaces Edition
Hugging Face Spaces allows you to have an interactive experience with the machine learning models, and we will be discovering the best application to get some inspiration.
Image by author
The top ten list is based on popularity, usability, and uniqueness. In this blog, we will be going to learn about the best machine learning demos on Hugging Face Spaces. The Spaces allow you to upload your Streamlit app, Gradio demo, and HTML application using Git.
This Pokémon Does Not Exist
This Pokémon Does Not Exist uses the ruDALL-E model to generate illustrations, and randomized names with attributes are chosen from the list. To collect the rare and unique pokemon, you need to write your name and press the submit button. This web app is simple but one of my favorites.
MAGMA (Multimodal Augmentation of Generative Models through Adapter-based Fine Tuning) is a Visual language model to describe or answer questions about the images. Learn more about MAGMA on arxiv.org. To use the ML Demo, you need to provide an image and ask a specific question. For example, “describe the image”. Learn more about use cases here.
AnimeGANv2 is the most popular machine learning application on Hugging Face Spaces with 515 ?. It also produces fast results with an unbelievable artistic touch. Learn more about interworking of generative models here. To use the demo, you need to upload a portrait and then choose the style to generate Anime-style art.
Image Restoration And Colorization
When I saw the Image Restoration and Colorization demo on Twitter, I thought they must be using a perfect example to show the outputs. But, when I tried it myself on a completely new photo, I was blown away by the simplicity and powerful functionality of the application. The Gradio demo asks you to upload the black&white and damaged image, and it will return a colored and high-quality photo. You can also play around with multiple options to get better results.
DiT Document Layout Analysis
DiT Document Layout Analysis demo uses a self-supervised pre-trained Document Image Transformer model to predict labels on a pdf document. For example, detecting tables, text, images, etc. The demo required a pdf document, and the rest is up to the powerful model to highlight various parts of the document.
Chef Transformer demo uses the t5-recipe-generation model to generate recipes based on chef, food style, and ingredients. If you are hungry and have limited food options, then type the ingredients and get the recipe for delicious food. This is my most favorite app as it is visually appealing with a unique use case.
ArcaneGAN Video uses flavored u-net trained on Arcane anime dataset, and images are generated via a blended stylegan2. Learn more about the model implementation here. For this Gradio demo, you just need to upload a sample video and let the model do the magic. The output video will be in Arcane anime style.
Rick & Morty ChatBot
Rick & Morty ChatBot uses a fine-tuned version of DialoGPT, which was trained on Rick and Morty's conversational dataset. The chatbot feature is new, and it provides you with an enhanced chat experience. Just type silly questions and keep the conversation going until you get bored.
OCR For Captcha
OCR For Captcha model was trained on a combination of CNN and RNN with an endpoint layer for implementing CTC loss. To learn in-depth about model training, check out Keras's code example. The app asked you to upload an image of a captcha and return highly accurate alphanumeric text.
Fastspeech2 TTS uses real-time state-of-the-art speech synthesis architectures such as Tacotron-2, Melgan, Multiband-Melgan, FastSpeech, FastSpeech2 based on TensorFlow. If you want a natural text-to-speech experience, try typing text and get amazed. This app also allows you to test from various model architectures. I just love the natural sound of a speaker.
When I was invited to the beta test of Spaces, I was skeptical, but within a month, the Spaces have outgrown Streamlit cloud, Heroku, and other cloud deployment platforms. In terms of ease of use, integrations, and faster inference. I am in love with HuggingFace Spaces and how community members are coming up with unique ideas for web applications. In this blog, we have covered the top ten list of machine learning demos on HF Spaces and learned how these applications work.
Please let me know your top ten ranking in the comments section.
Abid Ali Awan (@1abidaliawan) is a certified data scientist professional who loves building machine learning models. Currently, he is focusing on content creation and writing technical blogs on machine learning and data science technologies. Abid holds a Master's degree in Technology Management and a bachelor's degree in Telecommunication Engineering. His vision is to build an AI product using a graph neural network for students struggling with mental illness.