po-hsuan-huang Goto Github PK

api

Documentation and Samples for the Official HN API

awesome-deep-learning-papers

The most cited deep learning papers

awesome-tensorflow

TensorFlow - A curated list of dedicated resources http://tensorflow.org

awesome-visual-slam

:books: The list of vision-based SLAM / Visual Odometry open source, blogs, and papers

basic-yolo-keras

Easy training on custom dataset. Various backends (MobileNet and SqueezeNet) supported. A YOLO demo to detect raccoon run entirely in brower is accessible at https://git.io/vF7vI (not on Windows).

bundlefusion

BundleFusion: Real-time Globally Consistent 3D Reconstruction using Online Surface Re-integration

comparison-of-biophysically-plausible-implementation-of-neural-fields

Modeling of biophysically plausible neural networks in various scales has provided in sights in studies ranging from basic function of neural circuitry to mechanism of memory and sleep. This approach has been shown more promising than ever as more realistic models can be implemented thanks to the rapid advance of computer technologies. Nevertheless, model complexity and size still pose significant challenges to simulation speed and reproducibility. The simulation can be accelerated either by introducing concepts of software design, or by reduce the complexity of the model. Here, we demonstrate the computational utility is optimized by employing both strategies. We transport models of different neural levels from MatLab to NEST, and compare the results of simulation and the performance of the two software. On the other hand, we reduce the complexity of single neuron model and discuss the limitation of the simplified model. Finally, the computational speed is compared. This study shows NEST enables evaluation of the relations between psychophysical data and biophysical data by realizing implementation of complicated, large-scaled biophysically plausible neural fields .

deep-gaze-2

Deep Gaze II tensorflow reimplementation

deep-learning-papers-reading-roadmap

Deep Learning papers reading roadmap for anyone who are eager to learn this amazing tech!

deep-learning-study-group

deep-reinforcement-learning-papers

A list of recent papers regarding deep reinforcement learning

deepgaze

Computer Vision library for human-computer interaction. It implements Head Pose and Gaze Direction Estimation Using Convolutional Neural Networks, Skin Detection through Backprojection, Motion Detection and Tracking, Saliency Map.

diningtable

This product is in its infancy. We are building an Airbnb for restaurant business.

emoji-cheat-sheet.com

A one pager for emojis on Campfire and GitHub

enet

ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation

exploration-of-event-segmentation-theory-using-lstm

Publication : https://www.researchgate.net/publication/306323997_Exploration_of_event_segmentation_theory_using_LSTM Abstract: People tend to perceive ongoing continuous activity as a series of discrete events. Event Segmentation Theory (EST) postulates humans systematically partition continuous sensorimotor information flow into events and event boundaries (Reynolds, Zacks, and Braver, 2006). Gumbsch et.al. (Gumbsch and , 2016) investigated the basis of EST in the own motor interaction capabilities, and provided a computational model that learned events and event transitions while interacting with the environment. Their architecture uses a linear forward model as event models. We proposed that Long Short Term Memory (LSTM) neural network can augment the learning capability of forward models, and learn event transitions by forming gates. Gates are the cells of LSTM whose states switch on/off when the agent detects the event boundary. The computational model can then use the gates to predict event transitions and plan actions to achieve goals. We investigated the proper parameters for gate formation in a simple scenario. These parameters include length of buffer sequence, duration of training, and size of hidden layer. We also investigated the effect of weights, and found several classes of gates. We found weights on peephole to forget gate, input neuron to output gate, and cell output to output gate the defining weights of gate formation. They selectively open and close input gates, forget gates, and output gates at event boundaries. We also found output cell the most eligible candidate of gates for event boundary prediction. This findings can serve as guidelines of designing computational models that based on LSTM .

eye-movement-classification

Detect neurologically atypical behavior from eye movements

eye-tracking-saliency

Aligning eyetracking fixation points to the video frames and extract saliency values

face-resources

ffmpeg-cheat-sheet

ffmpeg cheat sheet specific for ezvisoin

ganhacks

starter from "How to Train a GAN?" at NIPS2016

high-throughput-realistic-data-augmentation-toolkit-for-opitical-digit-detection-on-fabrics

High Throughput, Realistic Data-Augmentation Toolkit for Opitical Digit Detection on Fabrics in Outdoors Enviroment

homework6

image-segmentation-project

[Computer Vision] Semantic Segmentation Demo with Tensorflow Keras

image_segmentation_project

jannlab-1

JANNLab Neural Network Framework for Java

keras-attention-mechanism

Attention mechanism Implementation for Keras.

mac_dotfiles

My Mac dot files

machine-learning-homework-nips-tuebingen

Homework of Machine learning course in 2015 winter semester

manutdzou.github.io