関連リンク
このページには研究しながら見つけた掘り出し物です。下記のリンクは全部私の研究と関係があります。他の人に僕のように役に立って欲しくて集めてきました。どうぞ閲覧してください。
機械学習
- dSP: Distributed structured prediction.
- LIBLINEAR: Linear SVM.
- LIBSVM: Non-linear SVM.
- Kaggle: Machine learning competitions.
- CMA-ES: Global black box function optimizer.
Deep Learning
フレームワーク
- Torch7 (github, cheatsheet): Flexible deep learning framework in Lua.
- Caffe (github, model zoo): Deep learning framework in C++ using model files.
- Theano (github): Deep learning framework in python that does symbolic compilation.
- TensorFlow (github): Google’s distributed deep learning framework.
実装
- Reinforcement Learning for Atari Games: Implementation of various approaches for reinforcement learning of atari games.
- Deep Residual Networks (torch: ILSVRC2015 winning network. Learns residual functions.
- Generative Adversial Networks (GAN) (torch): Image generation with deep networks.
- DecoupledNet: Semantic segmentation system.
- Neuraltalk (github): Image to text (Stanford).
- Show, Attend and Tell: Image to text (Toronto).
- Draw: Teaching network to draw.
- torch lstm (tree lstm: Long Short Term Memory unit implementation.
- R-CNN (Fast R-CNN, Faster R-CNN): Detection with convolutional networks.
- char-RNN (github): Character prediction with recursive neural networks.
- Learning to execute: Recursive neural network learns to generate python code.
- Deep Q-Network (Caffe, torch): Deep reinforcement learning for atari games.
- eyescream: Generating natural images with CNN.
- Deep Dreaming (github, torch, Caffe): Generating Neural Network art.
- DeepWalk (github): Deep learning for graphs.
- DeepPose: 2D pose estimation with CNN.
- Deep Filter Banks: Filter banks for texture recognition.
- Deep Convolutional Inverse Graphics Network (github): Encoder-decoder architecture that learns a graphics engine.
- Triplet Network: Learns from three examples simultaneously.
- Deconvolution Network for Segmentation (github): Encoder-decoder architectures for semantic segmentation.
- Deep Pink (github): Deep chess AI.
- Uncertainty in Deep Networks (github caffe, demos): Visualizing network uncertainty.
- Neural Artwork (github 1, github 2): Generating artwork with neural networks.
コンピュータビジョン
フレームワーク
- VLfeat (github): SIFT descriptors and other tools. Matlab/C interface.
- OpenCV: Large general usage computer vision library in C++.
顔認識等
- OpenFace: Full pipeline for face recognition using deep networks.
- IntraFace: Face attributes, feature detection, expression and expression transfer.
- VGG Face Descriptor: Deep network-based face descriptor.
ツール
- gPb superpixels: Large good quality superpixels.
- SLIC superpixels (opencv implementation): Fast regular superpixels.
- 9 superpixel algorithms (github): Benchmark of various superpixel algorithms.
- DAISY descriptor: Dense fast descriptor.
- Geodesic Object Proposals (3rd party github): Good object proposal approach.
- Image Specificity (github): Determine if an image is specific or not.
- SRCNN (github, waifu2x torch implementation): Super Resolution Convolutional Neural Network.
- Visual SfM: Frontend to some well known Structure from Motion (SfM) algorithms.
データセット
- Places (Places2): Scene classification dataset.
- Cross-Age Celebrity Dataset: Celebrity faces at different ages.
- CelebA dataset: Large amount of celebrity face images with landmark and attribute annotations.
- MovieBook and BookCorpus: Large book corpus and books aligned with movies.
- MPII Movie Description: Movies with caption dataset.
- Microsoft COCO: Segmentation, detection and text generation dataset.
- MPII Human Pose: 2D human pose estimation and activity dataset.
- KITTI: Autonomous driving related dataset (stereo, flow, odometry, detection, tracking, etc…).
- Fashionista: Clothes parsing dataset.
- Human3.6M: 3D human pose estimation dataset.
- Imagenet: Image classification dataset.
- CIFAR: 32x32 pixel image classification dataset.
- TinyImages: Really big 32x32 pixel image dataset.
面白い
- drawNet: See what’s going on in a network.
自然言語
- lda2vec: Modelling both word-to-word and document-to-word relationships jointly.
- Skip-thought vectors: Sentence encoding trained on books.
- sentence2vew (word2vec): Generate feature vectors from sentences or words.
- Stanford Sentiment Analysis: See how positive reviews are.
多様体
- manopt: General optimization on manifolds.
- smanifold: Exponential and logarithmic map for implicitly defined manifolds for principal geodesic analysis.
雑多
- LaTeX beamer mtheme: Good theme for presentations.
- The Elements of Style: Must read for good writing.