heatdrop.ai

Classic image processing and ML algorithms implemented in JavaScript, no GPU cluster required.

Companion code for the Chinese translation of "Programming Computer Vision with Python," collected in one repo so you don't have to type it out yourself.

A wrapper that wants you to graduate to raw OpenCV, not stay dependent on its abstractions.

A Python library that implements 1988-era Cellular Neural Networks for edge and corner detection—no GPUs required, just scipy and template math.

DeepLogo wraps TensorFlow's Object Detection API around the Flickr Logos 27 dataset, mostly as a working reference implementation.

A NumPy-native computer vision library that predates the deep-learning takeover and keeps chugging.

Video object tracking built from off-the-shelf parts for the ImageNet VID competition, with the rough edges left visible.

Turns screenshots of online chess games into FEN strings using a CNN, because manually transcribing boards is even more tedious than it sounds.

A TensorFlow 1.x reimplementation of the classic CRNN text-recognition architecture that trims 15% of parameters while nudging word error rate lower on standard synthetic data.

A research prototype that squeezes road segmentation, car detection, and street classification into a single real-time encoder-decoder model.

A TensorFlow implementation of the Single Shot MultiBox detector trained on stop signs and pedestrian crossings, with refreshingly honest notes about its limitations.

A CVPR 2017 face-landmark network whose reference implementation is now a time capsule of deep learning's past.

CFNet proves you don't need a hundred layers to follow a target through video—just a correlation filter trained end-to-end.

A straightforward TensorFlow 1.x implementation of the DenseNet paper, aimed at developers who want to see the architecture in code rather than equations.

A straightforward TensorFlow reimplementation of the densely connected convnet paper, built on someone else's ResNet code and benchmarked on standard CIFAR.

One Matlab repo that collects two decades of contrast enhancement algorithms so you don't have to hunt them down yourself.

HardNet trains image patch descriptors by deliberately mining the hardest negatives, not the closest misses.

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

An older TensorLayer reference implementation for BRATS brain tumor segmentation, with some honest notes about its own rough edges.

Unofficial Caffe port of Feature Pyramid Networks with VOC results and hard-won lessons about anchor scales.