public FPNFeatureExpander(HybridBlock network, string[] outputs, int[] num_filters, bool use_1x1 = true, bool use_upsample = true,
bool use_elewadd = true, bool use_p6 = false, bool p6_conv = true, bool no_bias = true,
bool pretrained = false, string norm_layer = null, FuncArgs norm_kwargs = null, Context ctx = null, string[] inputs = null, ParameterDict @params = null) : base(null, null, @params)
{
var(p_i, p_o, p_p) = __internal__.ParseNetwork(network, outputs, inputs, pretrained, ctx);
Construct(p_o, p_i, p_p, num_filters, use_1x1, use_upsample, use_elewadd, use_p6, p6_conv, no_bias, pretrained, norm_layer, norm_kwargs, ctx);
}
public static (Symbol, SymbolList, ParameterDict) ParseNetwork(HybridBlock network, string[] outputs, string[] inputs, bool pretrained, Context ctx)
{
var inputSymbols = inputs.Select(x => (Symbol.Var(x))).ToArray();
Symbol input = null;
if (inputSymbols.Length == 1)
{
input = inputSymbols[0];
}
else
{
input = Symbol.Group(inputSymbols);
}
ParameterDict @params = network.CollectParams();
var prefix = network.Prefix;
input = network.Call(input);
Debug.Assert(outputs.Length > 0, "At least one outputs must be specified.");
outputs = (from @out in outputs
select @out.EndsWith("_output") ? @out : @out + "_output").ToArray();
var outputSymbols = (from @out in outputs
select input[prefix + @out]).ToArray();
return(input, outputSymbols, @params);
}
public FeatureExtractor(HybridBlock network, string[] outputs, string[] inputs = null, bool pretrained = false, Context ctx = null, ParameterDict @params = null) : base(null, null, @params)
{
if (ctx == null)
{
ctx = mx.Cpu();
}
var(p_i, p_o, p_p) = __internal__.ParseNetwork(network, outputs, inputs, pretrained, ctx);
base.Construct(p_i, p_o, p_p);
}
public FasterRCNN(
HybridBlock features,
HybridBlock top_features,
string[] classes,
HybridBlock box_features = null,
int @short = 600,
int max_size = 1000,
int min_stage = 4,
int max_stage = 4,
string train_patterns = "",
float nms_thresh = 0.3f,
int nms_topk = 400,
int post_nms = 100,
string roi_mode = "align",
(int, int)?roi_size = null,
public BBoxArea(int axis = -1, string fmt = "corner", string prefix = "", ParameterDict @params = null) : base(prefix, @params)
{
if (fmt.ToLower() == "corner")
{
this._pre = new BBoxCornerToCenter(split: true);
}
else if (fmt.ToLower() == "center")
{
this._pre = new BBoxSplit(axis: axis);
}
else
{
throw new Exception($"Unsupported format: {fmt}. Use 'corner' or 'center'.");
}
}
public BBoxBatchIOU(int axis = -1, string fmt = "corner", int offset = 0, float eps = 1e-15f, string prefix = null, ParameterDict @params = null) : base(prefix, @params)
{
this._offset = offset;
this._eps = eps;
if (fmt.ToLower() == "center")
{
this._pre = new BBoxCenterToCorner(split: true);
}
else if (fmt.ToLower() == "corner")
{
this._pre = new BBoxSplit(axis: axis, squeeze_axis: true);
}
else
{
throw new Exception($"Unsupported format: {fmt}. Use 'corner' or 'center'.");
}
}
public FeatureExpander(HybridBlock network, string[] outputs, int[] num_filters, bool use_1x1_transition = true,
bool use_bn = true, float reduce_ratio = 1, int min_depth = 128, bool global_pool = false,
bool pretrained = false, Context ctx = null, string[] inputs = null, ParameterDict @params = null) : base(null, null, @params)
{
var(p_i, p_o, p_p) = __internal__.ParseNetwork(network, outputs, inputs, pretrained, ctx);
// append more layers
var y = p_o.Last();
var weight_init = new Xavier(rnd_type: "gaussian", factor_type: "out", magnitude: 2);
foreach (var _tup_2 in num_filters.Select((_p_1, _p_2) => Tuple.Create(_p_2, _p_1)))
{
var i = _tup_2.Item1;
var f = _tup_2.Item2;
if (use_1x1_transition)
{
var num_trans = Math.Max(min_depth, Convert.ToInt32(Math.Round(f * reduce_ratio)));
y = sym.Convolution(y, weight_init.InitWeight("weight"), null, num_filter: num_trans, kernel: new Shape(1, 1), no_bias: use_bn, symbol_name: $"expand_trans_conv{i}");
if (use_bn)
{
y = sym.BatchNorm(y, null, null, null, null, symbol_name: $"expand_trans_bn{i}");
}
y = sym.Activation(y, act_type: ActivationType.Relu, symbol_name: $"expand_trans_relu{i}");
}
y = sym.Convolution(y, weight_init.InitWeight("weight"), null, num_filter: f, kernel: new Shape(3, 3), pad: new Shape(1, 1), stride: new Shape(2, 2), no_bias: use_bn, symbol_name: $"expand_conv{i}");
if (use_bn)
{
y = sym.BatchNorm(y, null, null, null, null, symbol_name: $"expand_bn{i}");
}
y = sym.Activation(y, act_type: ActivationType.Relu, symbol_name: $"expand_reu{i}");
p_o.Add(y);
}
if (global_pool)
{
p_o.Add(sym.Pooling(y, pool_type: PoolingType.Avg, global_pool: true, kernel: new Shape(1, 1)));
}
base.Construct(p_o, p_i, p_p);
}
public FPNFeatureExpander(HybridBlock network, string[] outputs, int num_filters, bool use_1x1 = true, bool use_upsample = true,
bool use_elewadd = true, bool use_p6 = false, bool p6_conv = true, bool no_bias = true,
bool pretrained = false, HybridBlock norm_layer = null, FuncArgs norm_kwargs = null, Context ctx = null, string[] inputs = null, ParameterDict @params = null) : base(null, null, @params)
{
throw new NotImplementedException();
}
public AlphaPose(HybridBlock preact, int num_joints, string norm_layer = "BatchNorm", FuncArgs norm_kwargs = null, string prefix = null, ParameterDict @params = null) : base(prefix, @params)
{
throw new NotImplementedException();
}
public FeatureExpander(HybridBlock network, string[] outputs, int num_filters, bool use_1x1_transition = true,
bool use_bn = true, float reduce_ratio = 1, int min_depth = 128, bool global_pool = false,
bool pretrained = false, Context ctx = null, string[] inputs = null, ParameterDict @params = null) : base(null, null, @params)
{
}
private HybridSequential MakeLayer(HybridBlock block, int[] layers, int[] channels, int stride, int stage_index, int in_channels = 0, string norm_layer = "", bool last_gamma = false)
{
throw new NotImplementedException();
}
public static HybridSequential AddBlock(HybridSequential @out, HybridBlock block)
{
throw new NotImplementedException();
}
public Bottleneck(int planes, int inplanes, HybridBlock downsample = null, bool reduction = false, string norm_layer = "BatchNorm", FuncArgs kwargs = null, string prefix = "", ParameterDict @params = null) : base(prefix, @params)
{
throw new NotImplementedException();
}
public YOLO3DefaultTrainTransform(int width, int height, HybridBlock net = null, (float, float, float)?mean = null, (float, float, float)?std = null, float iou_thresh = 0.5f)
public MaskRCNN(HybridBlock features, HybridBlock top_features, string[] classes, int mask_channels = 256, int rcnn_max_dets = 1000, int rpn_test_pre_nms = 6000, int rpn_test_post_nms = 1000, int target_roi_scale = 1, int num_fcn_convs = 0, string norm_layer = "", FuncArgs norm_kwargs = null, string prefix = null, ParameterDict @params = null) : base(features: features, top_features: top_features, classes: classes, rpn_test_pre_nms: rpn_test_pre_nms, rpn_test_post_nms: rpn_test_post_nms, additional_output: true, prefix: prefix, @params: @params)
{
throw new NotImplementedException();
}
private HybridSequential MakeLayer(HybridBlock block, int[] layers, int[] channels, int stride, int stage_index, int in_channels = 0, string norm_layer = "BatchNorm", FuncArgs norm_kwargs = null)
{
throw new NotImplementedException();
}
public BottleneckV1b(int planes, int strides = 1, int dilation = 1, HybridBlock downsample = null, int previous_dilation = 1, string norm_layer = "", FuncArgs norm_kwargs = null, string prefix = null, ParameterDict @params = null) : base(prefix, @params)
{
throw new NotImplementedException();
}
public CenterNet(HybridBlock base_network, Dictionary <string, Dictionary <string, object> > heads, int classes, int head_conv_channel = 0, float scale = 4, int topk = 100, bool flip_test = false, int nms_thresh = 0, int nms_topk = 400, int post_nms = 100, string prefix = null, ParameterDict @params = null) : base(prefix, @params)
{
throw new NotImplementedException();
}
public RCNN(HybridBlock features, HybridBlock top_features, string[] classes, HybridBlock box_features, int @short, int max_size, string train_patterns,
float nms_thresh, int nms_topk, int post_nms, string roi_mode, (int, int) roi_size, (int, int) strides, float?clip, bool force_nms = false, string prefix = null, ParameterDict @params = null) : base(prefix, @params)
public SE_ResNetV1(HybridBlock block, int[] layers, int[] channels, int classes = 1000, bool thumbnail = false, string norm_layer = "BatchNorm", FuncArgs norm_kwargs = null, string prefix = null, ParameterDict @params = null) : base(prefix, @params)
{
throw new NotImplementedException();
}
public FeatureExtractor(HybridBlock network, string[] outputs, string[] inputs = null, bool pretrained = false, ParameterDict @params = null) : base(null, null, @params)
{
throw new NotImplementedException();
}
public FasterRCNNTrainBatchify(HybridBlock net, int num_shards = 1)
{
throw new NotImplementedException();
}
public FasterRCNNDefaultTrainTransform(int @short = 600, int max_size = 1000, HybridBlock net = null, (float, float, float)?mean = null, (float, float, float)?std = null, (float, float, float)?box_norm = null, int num_sample = 256, float pos_iou_thresh = 0.7f, float neg_iou_thresh = 0.3f, float pos_ratio = 0.5f, float flip_p = 0.5f, int ashape = 128, bool multi_stage = false)
public CIFARResNetV2(HybridBlock block, int[] layers, int[] channels, int classes = 10, string norm_layer = "BatchNorm", FuncArgs norm_kwargs = null, string prefix = "", ParameterDict @params = null) : base(prefix, @params)
{
throw new NotImplementedException();
}
public static void ExportBlock(string path, HybridBlock block, Shape data_shape = null, int epoch = 0, bool preprocess = true, string layout = "HWC", Context ctx = null)
{
throw new NotImplementedException();
}
public Tree(int levels, HybridBlock block, int in_channels, int out_channels, int stride = 1, bool level_root = false, int root_dim = 0, int root_kernel_size = 1,
int dilation = 1, bool root_residual = false, string norm_layer = "BatchNorm", FuncArgs norm_kwargs = null, string prefix = null, ParameterDict @params = null) : base(prefix, @params)
{
throw new NotImplementedException();
}
public static (SymbolList, SymbolList, ParameterDict) ParseNetwork(HybridBlock network, string[] outputs, string[] inputs, bool pretrained, Context ctx)
{
throw new NotImplementedException();
}
public HybridSequential MakeLayer(HybridBlock block, int planes, int blocks, int stride = 1)
{
throw new NotImplementedException();
}
public static void ExportBlock(string path, HybridBlock block, Shape data_shape = null, int epoch = 0, HybridBlock preprocess = null, string layout = "HWC", Context ctx = null)
{
if (ctx == null)
{
ctx = mx.Cpu();
}
NDArray x;
int t;
int c;
int w;
int h;
HybridSequential wrapper_block;
List <Shape> data_shapes = new List <Shape>();
// input image layout
layout = layout.ToUpper();
if (data_shape == null)
{
if (layout == "HWC")
{
data_shapes = (from s in new int[] { 224, 256, 299, 300, 320, 416, 512, 600 }
select new Shape(s, s, 3)).ToList();
}
else if (layout == "CHW")
{
data_shapes = (from s in new int[] { 224, 256, 299, 300, 320, 416, 512, 600 }
select new Shape(3, s, s)).ToList();
}
else
{
throw new Exception("Unable to predict data_shape, please specify.");
}
}
else
{
data_shapes.Add(data_shape);
}
if (preprocess != null)
{
wrapper_block = new HybridSequential();
preprocess.Initialize(ctx: new Context[] { ctx });
wrapper_block.Add(preprocess);
wrapper_block.Add(block);
}
else
{
wrapper_block = new HybridSequential();
preprocess.Initialize(ctx: new Context[] { ctx });
wrapper_block.Add(block);
Debug.Assert(new string[] { "CHW", "CTHW" }.Contains(layout), $"Default layout is CHW for 2D models and CTHW for 3D models if preprocess is None, provided {layout}.");
}
wrapper_block.CollectParams().ResetCtx(ctx);
// try different data_shape if possible, until one fits the network
object last_exception = null;
foreach (var dshape in data_shapes)
{
if (layout == "HWC")
{
(h, w, c) = dshape;
x = nd.Zeros(new Shape(1, h, w, c), ctx: ctx);
}
else if (layout == "CHW")
{
(c, h, w) = dshape;
x = nd.Zeros(new Shape(1, c, h, w), ctx: ctx);
}
else if (layout == "THWC")
{
(t, h, w, c) = dshape;
x = nd.Zeros(new Shape(1, t, h, w, c), ctx: ctx);
}
else if (layout == "CTHW")
{
(c, t, h, w) = dshape;
x = nd.Zeros(new Shape(1, c, t, h, w), ctx: ctx);
}
else
{
throw new Exception(String.Format("Input layout {0} is not supported yet.", layout));
}
// hybridize and forward once
wrapper_block.Hybridize();
try
{
wrapper_block.Call(x);
wrapper_block.Export(path, epoch);
last_exception = null;
break;
}
catch (MXNetException ex)
{
last_exception = ex;
}
}
if (last_exception != null)
{
throw new Exception(last_exception.ToString().Split('\n')[0]);
}
}