smile.deep.layer.GroupNormLayer Maven / Gradle / Ivy

Go to download
Show more of this group Show more artifacts with this name
Show all versions of smile-deep Show documentation
smile-deep
The newest version!
/*
 * Copyright (c) 2010-2024 Haifeng Li. All rights reserved.
 *
 * Smile is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * Smile is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with Smile.  If not, see .
 */
package smile.deep.layer;

import org.bytedeco.pytorch.GroupNormImpl;
import org.bytedeco.pytorch.GroupNormOptions;
import org.bytedeco.pytorch.Module;
import smile.deep.tensor.Tensor;

/**
 * Group normalization. The input channels are separated into groups.
 * The mean and standard-deviation are calculated separately over each
 * group.
 *
 * @author Haifeng Li
 */
public class GroupNormLayer implements Layer {
    /** Implementation. */
    private final GroupNormImpl module;

    /**
     * Constructor.
     * @param groups the number of groups to separate the channels into.
     *               The number of channels must be divisible by the number
     *               of groups.
     * @param channels the number of input channels in (N,C,H,W).
     */
    public GroupNormLayer(int groups, int channels) {
        this(groups, channels, 1E-05, true);
    }

    /**
     * Constructor.
     * @param groups the number of groups to separate the channels into.
     *               The number of channels must be divisible by the number
     *               of groups.
     * @param channels the number of input channels in (N,C,H,W).
     * @param eps a value added to the denominator for numerical stability.
     * @param affine when set to true, this layer has learnable affine parameters.
     */
    public GroupNormLayer(int groups, int channels, double eps, boolean affine) {
        var options = new GroupNormOptions(groups, channels);
        options.eps().put(eps);
        options.affine().put(affine);
        this.module = new GroupNormImpl(options);
    }

    @Override
    public Module asTorch() {
        return module;
    }

    @Override
    public Tensor forward(Tensor input) {
        return new Tensor(module.forward(input.asTorch()));
    }
}