VelesForge

World first! Number one!

Dr. Yong Min Lee.

Register

AlexNet master

Model created for object recognition. Database - Imagenet

Author: VELES team
Updated: 2015-11-23 09:42:37
Details:
Model created for object recognition. Database - Imagenet (1000 classes). Self-constructing Model. It means that Model can change for any Model (Convolutional, Fully connected, different parameters) in configuration file.

Approximator master

Model created for functions approximation.

Author: VELES team
Updated: 2015-11-23 09:10:55
Details:
Model created for functions approximation. Dataset - matlab files with x and y points. Model - fully-connected Neural Network with MSE loss function.

CIFAR10 master

Model created for object recognition with CIFAR10 dataset.

Author: VELES team
Updated: 2015-11-23 10:01:01
Details:
Model created for object recognition. Dataset - CIFAR10. Self-constructing Model. It means that Model can change for any Model (Convolutional, Fully connected, different parameters) in configuration file. Package contains two configuration files: cifar_caffe_config for Convolutional Neural Network with parameters just like in Caffe and cifar_config for Fully-connected Neural Network.

DemoKohonen master

Kohonen map demo on a simple two dimension dataset.

Author: VELES team
Updated: 2015-11-23 10:02:23
Details:
Kohonen map demo on a simple two dimension dataset. Package contains dataset.

Hands master

Model created for human hands recognition.

Author: VELES team
Updated: 2015-11-23 10:03:24
Details:
Model created for human hands recognition. Dataset - Samsung Database with images of human hands. Model - fully-connected Neural Network with SoftMax loss function.

ImagenetAE master

CNN with pretraining of each layer by Autoencoder created for object recognition on Imagenet dataset

Author: VELES team
Updated: 2015-11-23 10:03:46
Details:
Model created for object recognition. Dataset - Imagenet (DET challenge). Model - convolutional neural network, dynamically constructed, with pretraining of all layers one by one with autoencoder.

Kanji master

Model created for Chinese characters recognition

Author: VELES team
Updated: 2015-11-23 10:19:20
Details:
Model created for Chinese characters recognition. Dataset was generated by VELES with generate_kanji.py utility. Self-constructing Model. It means that Model can change for any Model (Convolutional, Fully connected, different parameters) in configuration file. Current model - fully-connected Neural Network with MSE loss function.

Lines master

Model created for geometric figure recognition.

Author: VELES team
Updated: 2015-11-23 10:31:09
Details:
Model created for geometric figure recognition. Dataset was synthetically generated by VELES. Self-constructing Model. It means that Model can change for any Model (Convolutional, Fully connected, different parameters) in configuration file. Current model - Convolutional Neural Network.

MNIST master

Model created for digits recognition with MNIST database.

Author: VELES team
Updated: 2015-11-23 10:37:41
Details:
Model created for digits recognition. Database - MNIST. Self-constructing Model. It means that Model can change for any Model (Convolutional, Fully connected, different parameters) in configuration file. Package contains three configuration files: mnist_config for fully-connected Neural Network with parameters, which can be modified by Genetic algorithm, mnist_conv_config for convolutional Neural Network with parameters, which were modified by Genetic Algorithm, mnist_caffe_config for convolutional Neural Network with parameters just like in Caffe.

Mnist7 master

Model created for digits recognition with MNIST Database with target encoded as 7 points.

Author: VELES team
Updated: 2015-11-23 10:40:50
Details:
Model created for digits recognition. Database - MNIST. Model - fully-connected Neural Network with MSE loss function with target encoded as 7 points.

MnistAE master

Model created for digits recognition with MNIST Database by autoencoder.

Author: VELES team
Updated: 2015-11-23 10:43:49
Details:
Model created for digits recognition. Database - MNIST. Model - autoencoder.

MnistSimple master

Model created for digits recognition with MNIST database by simple fully-connected neural network.

Author: VELES team
Updated: 2015-11-23 10:46:38
Details:
Model created for digits recognition. Database - MNIST. Model - fully-connected Neural Network with SoftMax loss function.

SpamKohonen master

Kohonen Spam detection on Lee Man Ha dataset.

Author: VELES team
Updated: 2015-11-23 10:56:29
Details:
Kohonen Spam detection on Lee Man Ha dataset.

Stl10 master

Model for object recognition with STL10 database.

Author: VELES Team
Updated: 2015-11-23 11:02:42
Details:
Self-constructing Workflow (Fully-connected/Convolutional) for recognition of objects on base of STL-10 data.

TvChannels master

Model created for object recognition (logotypes of TV channels)

Author: VELES team
Updated: 2015-11-23 11:14:19
Details:
Model created for object recognition (logotypes of TV channels). Dataset - Channels. Self-constructing Model. It means that Model can change for any Model (Convolutional, Fully connected, different parameters) in configuration file. You can find boxer, which automaticly find logotypes and generate dataset, in external file

VideoAE master

Model created for compress video with autoencoder

Author: VELES team
Updated: 2015-11-23 11:17:58
Details:
Model created for compress video. Model - autoencoder.

Wine master

Model created for class of wine recognition

Author: Veles team
Updated: 2015-11-23 11:21:30
Details:
Model created for class of wine recognition. Database - Wine. Model - fully-connected Neural Network with SoftMax loss function. Package contains dataset.

YaleFaces master

Model was created for face recognition.

Author: VELES team
Updated: 2015-11-23 11:26:49
Details:
Model was created for face recognition. Database - Yale Faces. Self-constructing Model. It means that Model can change for any Model (Convolutional, Fully connected, different parameters) in configuration file. Current model - fully-connected Neural Network with SoftMax loss function. Also you can find preprocessing workflow for yale faces dataset