File: AutoFormerV2\ConvLayer.cs
Web Access
Project: src\src\Microsoft.ML.TorchSharp\Microsoft.ML.TorchSharp.csproj (Microsoft.ML.TorchSharp)
// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.
// See the LICENSE file in the project root for more information.
 
using TorchSharp;
using TorchSharp.Modules;
using static TorchSharp.torch;
using static TorchSharp.torch.nn;
 
namespace Microsoft.ML.TorchSharp.AutoFormerV2
{
    /// <summary>
    /// The layer of convolution blocks in AutoFormer network.
    /// </summary>
    public class ConvLayer : Module<Tensor, int, int, (Tensor, int, int, Tensor, int, int)>
    {
#pragma warning disable MSML_PrivateFieldName // Need to match TorchSharp model names.
        private readonly ModuleList<MBConv> blocks;
        private readonly PatchMerging downsample;
        private bool _disposedValue;
#pragma warning restore MSML_PrivateFieldName
 
        /// <summary>
        /// Initializes a new instance of the <see cref="ConvLayer"/> class.
        /// </summary>
        /// <param name="inChannels">The input channels.</param>
        /// <param name="outChannels">The output channels.</param>
        /// <param name="depth">The number of blocks.</param>
        /// <param name="convExpandRatio">The expand ratio of convolution layer.</param>
        public ConvLayer(int inChannels, int outChannels, int depth, double convExpandRatio = 4.0)
            : base(nameof(ConvLayer))
        {
            this.blocks = new ModuleList<MBConv>();
            for (int i = 0; i < depth; i++)
            {
                this.blocks.Add(new MBConv(inChannels, inChannels, convExpandRatio));
            }
 
            this.downsample = new PatchMerging(inChannels, outChannels);
        }
 
        /// <inheritdoc/>
        [System.Diagnostics.CodeAnalysis.SuppressMessage("Naming", "MSML_GeneralName:This name should be PascalCased", Justification = "Need to match TorchSharp.")]
        public override (Tensor, int, int, Tensor, int, int) forward(Tensor x, int h, int w)
        {
            using (var scope = torch.NewDisposeScope())
            {
                foreach (var block in this.blocks)
                {
                    x = block.forward(x);
                }
 
                var xOut = x;
                var (xTmp, nH, nW) = this.downsample.forward(x, h, w);
                x = xTmp;
 
                return (xOut.MoveToOuterDisposeScope(), h, w, x.MoveToOuterDisposeScope(), nH, nW);
            }
        }
 
        protected override void Dispose(bool disposing)
        {
            if (!_disposedValue)
            {
                if (disposing)
                {
                    blocks.Dispose();
                    downsample.Dispose();
                    _disposedValue = true;
                }
            }
 
            base.Dispose(disposing);
        }
    }
}