August 2017 Summaries

Ngrok is a reverse proxy tool that enables developers to showcase their locally running web services to others without deploying them to a server, making it highly useful for demonstrating projects at hackathons or sharing app progress with clients. The process involves downloading, unzipping, and launching ngrok for your operating system, which is demonstrated for both macOS/Linux and Windows users. To use ngrok, you need a locally running server, which can be set up with a simple Sinatra "Hello, World" app, and then ngrok exposes this service by specifying the protocol and port, with Sinatra defaulting to port 4567. This straightforward setup allows developers to share their work quickly and effortlessly.

The text introduces a new demo for a visual recognition API that allows users to test image and video tagging capabilities. This updated demo offers an enhanced, user-friendly interface that supports multiple media file testing, enabling users to save and compare results from different recognition models. Users can also set their own confidence levels for predicted tags and access the demo in over 20 languages, with the system recognizing over 11,000 different concepts. The demo highlights the impressive video recognition feature, which identifies tags in specific video segments, and invites users to try the technology themselves to experience its capabilities firsthand.

Mike Knowles, a lead developer at Photobucket, leveraged Clarifai's machine learning technology to enhance the platform's content moderation capabilities, significantly improving the efficiency and effectiveness of managing user-generated content (UGC). Photobucket, a major image and video hosting service, faced challenges in moderating the vast amounts of content uploaded daily, which posed risks of illegal or offensive material slipping through manual checks by human moderators. By integrating Clarifai’s Not Safe for Work (NSFW) nudity recognition model, Photobucket achieved real-time moderation of all uploaded images, increasing the detection of inappropriate content from a mere 0.1% to 70%. This automation allowed for a reduction in the human moderation team, with staff transitioning to customer support roles, thereby enhancing overall user experience. Additionally, the implementation, completed in just 12 weeks, enabled Photobucket to proactively address harmful content, even leading to the identification of child pornography accounts. Mike chose Clarifai over other options due to its superior accuracy, ease of use, and comprehensive support, ultimately making the internet a safer environment while mitigating Photobucket's legal and financial risks.

The new Face Embedding model by Clarifai enables users to organize, filter, and rank facial images based on visual similarity by transforming them into numerical vectors in a 1024-dimensional space. This model, derived from the Face Detection model, measures visual similarity by calculating the distance between these vectors, with closer vectors indicating greater similarity. The provided tutorial demonstrates how to build a simple application that compares facial similarity between parents and their child using Python and the Numpy library. The process involves initializing the Clarifai app, obtaining embeddings for each face in the images, and calculating the distances between them to determine which parent the child resembles more. The tutorial concludes by encouraging users to experiment with their images and explore other potential applications of face embeddings, such as facial recognition-based searches and deduplication of similar images.