search

Daniel-Liu-c0deb0t / Java-Machine-Learning

Deep learning library for Java, with fully connected, convolutional, and recurrent layers. Also features many gradient descent optimization algorithms.
135 stars 25 forks source link

Make image classification model (like a tutorial) #9

Open HanbumKo opened 5 years ago

HanbumKo commented 5 years ago

Now, tutorial codes using MNIST. We need to add actual classification examples

Daniel-Liu-c0deb0t commented 5 years ago

I think the current convolution layer supports 3 color channels, so not much work needs to be done in terms of updating the library. There was a bug fix that should have lowered the training time by a lot, so classifying larger datasets of 2D color images (maybe a partial CIFAR-10) is feasible. Another way to avoid training and quickly get the weights would be to add code to read pretrained weights from other libraries. That may be much faster than training on a CPU.

Also, another dataset to consider is the sign language dataset, which is similar to MNIST.

HanbumKo commented 5 years ago

I think so. Someone who knows other library's format of weight saved file can implement the concept.