heatdrop.ai

A Node.js wrapper for running pre-trained TensorFlow graphs without installing Python on your production servers.

A reading list that tracks how to compress models and cram them onto FPGAs/ASICs without the marketing fluff.

A thin Python wrapper around Stanford's Java CoreNLP server, now officially abandoned in favor of Stanza.

A weekend project that bolts deep_sort and SORT onto darkflow so you can follow people through video without writing the integration yourself.

A PHP extension that wraps OpenCV's computer-vision library so you can do face recognition and image processing without leaving your comfort language.

CERN's dist-keras wraps Keras models in Apache Spark to run data-parallel deep learning across clusters, with a research-friendly focus on pluggable distributed optimizers.

A Cython wrapper that lets you pip-install YOLO v3 instead of wrestling Darknet's Makefile.

A no-frills Python wrapper that pipes webcam or video files through TensorFlow object detection, containerized for Linux purists.

A minimal C++ inference engine for Keras networks, built for when you need predictions but can't ship a Python interpreter.

A JavaScript wrapper that squeezes Tiny YOLO into TensorFlow.js so your webcam can (slowly) find cats without a server round-trip.

A resource-sipping matrix-math coprocessor in VHDL that brings Google's TPU architecture down to FPGA scale.

Google's real-time pose estimation model, originally TensorFlow.js, got dragged into Python-land and then partially rescued from JavaScript's performance sins.

A CTC decoder that constrains output to dictionary words while gracefully handling numbers, punctuation, and other non-word characters that pure token passing chokes on.

An early, unvarnished port of real-time object detection to Android that shows exactly how much heavy lifting "mobile ML" used to require.

A 2019 recipe for running heavy object detection on edge hardware by offloading inference to Intel's Neural Compute Stick.

A grab-bag of GPU-accelerated vision models you can compile and run on hardware from Jetson down to Raspberry Pi.

A stripped-down C++ inference engine that trades framework overhead for raw CUDA and MKL speed.

A demo that wrangles camera frames into TensorFlow.js inside China's biggest super-app, after years of fighting platform quirks.

A compatibility layer that compiles CUDA code to OpenCL 1.2, so TensorFlow runs on Radeon, Intel, and older hardware.

A dead-simple visual search engine that extracts VGG16 features and serves them over Flask, built to teach rather than scale.