add wb

models/experimental.py

@@ -0,0 +1,262 @@
+import numpy as np
+import random
+import torch
+import torch.nn as nn
+
+from models.common import Conv, DWConv
+from utils.google_utils import attempt_download
+
+
+class CrossConv(nn.Module):
+    # Cross Convolution Downsample
+    def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False):
+        # ch_in, ch_out, kernel, stride, groups, expansion, shortcut
+        super(CrossConv, self).__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, (1, k), (1, s))
+        self.cv2 = Conv(c_, c2, (k, 1), (s, 1), g=g)
+        self.add = shortcut and c1 == c2
+
+    def forward(self, x):
+        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
+
+
+class Sum(nn.Module):
+    # Weighted sum of 2 or more layers https://arxiv.org/abs/1911.09070
+    def __init__(self, n, weight=False):  # n: number of inputs
+        super(Sum, self).__init__()
+        self.weight = weight  # apply weights boolean
+        self.iter = range(n - 1)  # iter object
+        if weight:
+            self.w = nn.Parameter(-torch.arange(1., n) / 2, requires_grad=True)  # layer weights
+
+    def forward(self, x):
+        y = x[0]  # no weight
+        if self.weight:
+            w = torch.sigmoid(self.w) * 2
+            for i in self.iter:
+                y = y + x[i + 1] * w[i]
+        else:
+            for i in self.iter:
+                y = y + x[i + 1]
+        return y
+
+
+class MixConv2d(nn.Module):
+    # Mixed Depthwise Conv https://arxiv.org/abs/1907.09595
+    def __init__(self, c1, c2, k=(1, 3), s=1, equal_ch=True):
+        super(MixConv2d, self).__init__()
+        groups = len(k)
+        if equal_ch:  # equal c_ per group
+            i = torch.linspace(0, groups - 1E-6, c2).floor()  # c2 indices
+            c_ = [(i == g).sum() for g in range(groups)]  # intermediate channels
+        else:  # equal weight.numel() per group
+            b = [c2] + [0] * groups
+            a = np.eye(groups + 1, groups, k=-1)
+            a -= np.roll(a, 1, axis=1)
+            a *= np.array(k) ** 2
+            a[0] = 1
+            c_ = np.linalg.lstsq(a, b, rcond=None)[0].round()  # solve for equal weight indices, ax = b
+
+        self.m = nn.ModuleList([nn.Conv2d(c1, int(c_[g]), k[g], s, k[g] // 2, bias=False) for g in range(groups)])
+        self.bn = nn.BatchNorm2d(c2)
+        self.act = nn.LeakyReLU(0.1, inplace=True)
+
+    def forward(self, x):
+        return x + self.act(self.bn(torch.cat([m(x) for m in self.m], 1)))
+
+
+class Ensemble(nn.ModuleList):
+    # Ensemble of models
+    def __init__(self):
+        super(Ensemble, self).__init__()
+
+    def forward(self, x, augment=False):
+        y = []
+        for module in self:
+            y.append(module(x, augment)[0])
+        # y = torch.stack(y).max(0)[0]  # max ensemble
+        # y = torch.stack(y).mean(0)  # mean ensemble
+        y = torch.cat(y, 1)  # nms ensemble
+        return y, None  # inference, train output
+
+
+
+
+
+class ORT_NMS(torch.autograd.Function):
+    '''ONNX-Runtime NMS operation'''
+    @staticmethod
+    def forward(ctx,
+                boxes,
+                scores,
+                max_output_boxes_per_class=torch.tensor([100]),
+                iou_threshold=torch.tensor([0.45]),
+                score_threshold=torch.tensor([0.25])):
+        device = boxes.device
+        batch = scores.shape[0]
+        num_det = random.randint(0, 100)
+        batches = torch.randint(0, batch, (num_det,)).sort()[0].to(device)
+        idxs = torch.arange(100, 100 + num_det).to(device)
+        zeros = torch.zeros((num_det,), dtype=torch.int64).to(device)
+        selected_indices = torch.cat([batches[None], zeros[None], idxs[None]], 0).T.contiguous()
+        selected_indices = selected_indices.to(torch.int64)
+        return selected_indices
+
+    @staticmethod
+    def symbolic(g, boxes, scores, max_output_boxes_per_class, iou_threshold, score_threshold):
+        return g.op("NonMaxSuppression", boxes, scores, max_output_boxes_per_class, iou_threshold, score_threshold)
+
+
+class TRT_NMS(torch.autograd.Function):
+    '''TensorRT NMS operation'''
+    @staticmethod
+    def forward(
+        ctx,
+        boxes,
+        scores,
+        background_class=-1,
+        box_coding=1,
+        iou_threshold=0.45,
+        max_output_boxes=100,
+        plugin_version="1",
+        score_activation=0,
+        score_threshold=0.25,
+    ):
+        batch_size, num_boxes, num_classes = scores.shape
+        num_det = torch.randint(0, max_output_boxes, (batch_size, 1), dtype=torch.int32)
+        det_boxes = torch.randn(batch_size, max_output_boxes, 4)
+        det_scores = torch.randn(batch_size, max_output_boxes)
+        det_classes = torch.randint(0, num_classes, (batch_size, max_output_boxes), dtype=torch.int32)
+        return num_det, det_boxes, det_scores, det_classes
+
+    @staticmethod
+    def symbolic(g,
+                 boxes,
+                 scores,
+                 background_class=-1,
+                 box_coding=1,
+                 iou_threshold=0.45,
+                 max_output_boxes=100,
+                 plugin_version="1",
+                 score_activation=0,
+                 score_threshold=0.25):
+        out = g.op("TRT::EfficientNMS_TRT",
+                   boxes,
+                   scores,
+                   background_class_i=background_class,
+                   box_coding_i=box_coding,
+                   iou_threshold_f=iou_threshold,
+                   max_output_boxes_i=max_output_boxes,
+                   plugin_version_s=plugin_version,
+                   score_activation_i=score_activation,
+                   score_threshold_f=score_threshold,
+                   outputs=4)
+        nums, boxes, scores, classes = out
+        return nums, boxes, scores, classes
+
+
+class ONNX_ORT(nn.Module):
+    '''onnx module with ONNX-Runtime NMS operation.'''
+    def __init__(self, max_obj=100, iou_thres=0.45, score_thres=0.25, max_wh=640, device=None):
+        super().__init__()
+        self.device = device if device else torch.device("cpu")
+        self.max_obj = torch.tensor([max_obj]).to(device)
+        self.iou_threshold = torch.tensor([iou_thres]).to(device)
+        self.score_threshold = torch.tensor([score_thres]).to(device)
+        self.max_wh = max_wh # if max_wh != 0 : non-agnostic else : agnostic
+        self.convert_matrix = torch.tensor([[1, 0, 1, 0], [0, 1, 0, 1], [-0.5, 0, 0.5, 0], [0, -0.5, 0, 0.5]],
+                                           dtype=torch.float32,
+                                           device=self.device)
+
+    def forward(self, x):
+        boxes = x[:, :, :4]
+        conf = x[:, :, 4:5]
+        scores = x[:, :, 5:]
+        scores *= conf
+        boxes @= self.convert_matrix
+        max_score, category_id = scores.max(2, keepdim=True)
+        dis = category_id.float() * self.max_wh
+        nmsbox = boxes + dis
+        max_score_tp = max_score.transpose(1, 2).contiguous()
+        selected_indices = ORT_NMS.apply(nmsbox, max_score_tp, self.max_obj, self.iou_threshold, self.score_threshold)
+        X, Y = selected_indices[:, 0], selected_indices[:, 2]
+        selected_boxes = boxes[X, Y, :]
+        selected_categories = category_id[X, Y, :].float()
+        selected_scores = max_score[X, Y, :]
+        X = X.unsqueeze(1).float()
+        return torch.cat([X, selected_boxes, selected_categories, selected_scores], 1)
+
+class ONNX_TRT(nn.Module):
+    '''onnx module with TensorRT NMS operation.'''
+    def __init__(self, max_obj=100, iou_thres=0.45, score_thres=0.25, max_wh=None ,device=None):
+        super().__init__()
+        assert max_wh is None
+        self.device = device if device else torch.device('cpu')
+        self.background_class = -1,
+        self.box_coding = 1,
+        self.iou_threshold = iou_thres
+        self.max_obj = max_obj
+        self.plugin_version = '1'
+        self.score_activation = 0
+        self.score_threshold = score_thres
+
+    def forward(self, x):
+        boxes = x[:, :, :4]
+        conf = x[:, :, 4:5]
+        scores = x[:, :, 5:]
+        scores *= conf
+        num_det, det_boxes, det_scores, det_classes = TRT_NMS.apply(boxes, scores, self.background_class, self.box_coding,
+                                                                    self.iou_threshold, self.max_obj,
+                                                                    self.plugin_version, self.score_activation,
+                                                                    self.score_threshold)
+        return num_det, det_boxes, det_scores, det_classes
+
+
+class End2End(nn.Module):
+    '''export onnx or tensorrt model with NMS operation.'''
+    def __init__(self, model, max_obj=100, iou_thres=0.45, score_thres=0.25, max_wh=None, device=None):
+        super().__init__()
+        device = device if device else torch.device('cpu')
+        assert isinstance(max_wh,(int)) or max_wh is None
+        self.model = model.to(device)
+        self.model.model[-1].end2end = True
+        self.patch_model = ONNX_TRT if max_wh is None else ONNX_ORT
+        self.end2end = self.patch_model(max_obj, iou_thres, score_thres, max_wh, device)
+        self.end2end.eval()
+
+    def forward(self, x):
+        x = self.model(x)
+        x = self.end2end(x)
+        return x
+
+
+
+
+
+def attempt_load(weights, map_location=None):
+    # Loads an ensemble of models weights=[a,b,c] or a single model weights=[a] or weights=a
+    model = Ensemble()
+    for w in weights if isinstance(weights, list) else [weights]:
+        attempt_download(w)
+        ckpt = torch.load(w, map_location=map_location)  # load
+        model.append(ckpt['ema' if ckpt.get('ema') else 'model'].float().fuse().eval())  # FP32 model
+    
+    # Compatibility updates
+    for m in model.modules():
+        if type(m) in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU]:
+            m.inplace = True  # pytorch 1.7.0 compatibility
+        elif type(m) is nn.Upsample:
+            m.recompute_scale_factor = None  # torch 1.11.0 compatibility
+        elif type(m) is Conv:
+            m._non_persistent_buffers_set = set()  # pytorch 1.6.0 compatibility
+    
+    if len(model) == 1:
+        return model[-1]  # return model
+    else:
+        print('Ensemble created with %s\n' % weights)
+        for k in ['names', 'stride']:
+            setattr(model, k, getattr(model[-1], k))
+        return model  # return ensemble
+
+

models/export.py

@@ -0,0 +1,98 @@
+import argparse
+import sys
+import time
+
+sys.path.append('./')  # to run '$ python *.py' files in subdirectories
+
+import torch
+import torch.nn as nn
+
+import models
+from models.experimental import attempt_load
+from utils.activations import Hardswish, SiLU
+from utils.general import set_logging, check_img_size
+from utils.torch_utils import select_device
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--weights', type=str, default='./yolor-csp-c.pt', help='weights path')
+    parser.add_argument('--img-size', nargs='+', type=int, default=[640, 640], help='image size')  # height, width
+    parser.add_argument('--batch-size', type=int, default=1, help='batch size')
+    parser.add_argument('--dynamic', action='store_true', help='dynamic ONNX axes')
+    parser.add_argument('--grid', action='store_true', help='export Detect() layer grid')
+    parser.add_argument('--device', default='cpu', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
+    opt = parser.parse_args()
+    opt.img_size *= 2 if len(opt.img_size) == 1 else 1  # expand
+    print(opt)
+    set_logging()
+    t = time.time()
+
+    # Load PyTorch model
+    device = select_device(opt.device)
+    model = attempt_load(opt.weights, map_location=device)  # load FP32 model
+    labels = model.names
+
+    # Checks
+    gs = int(max(model.stride))  # grid size (max stride)
+    opt.img_size = [check_img_size(x, gs) for x in opt.img_size]  # verify img_size are gs-multiples
+
+    # Input
+    img = torch.zeros(opt.batch_size, 3, *opt.img_size).to(device)  # image size(1,3,320,192) iDetection
+
+    # Update model
+    for k, m in model.named_modules():
+        m._non_persistent_buffers_set = set()  # pytorch 1.6.0 compatibility
+        if isinstance(m, models.common.Conv):  # assign export-friendly activations
+            if isinstance(m.act, nn.Hardswish):
+                m.act = Hardswish()
+            elif isinstance(m.act, nn.SiLU):
+                m.act = SiLU()
+        # elif isinstance(m, models.yolo.Detect):
+        #     m.forward = m.forward_export  # assign forward (optional)
+    model.model[-1].export = not opt.grid  # set Detect() layer grid export
+    y = model(img)  # dry run
+
+    # TorchScript export
+    try:
+        print('\nStarting TorchScript export with torch %s...' % torch.__version__)
+        f = opt.weights.replace('.pt', '.torchscript.pt')  # filename
+        ts = torch.jit.trace(model, img, strict=False)
+        ts.save(f)
+        print('TorchScript export success, saved as %s' % f)
+    except Exception as e:
+        print('TorchScript export failure: %s' % e)
+
+    # ONNX export
+    try:
+        import onnx
+
+        print('\nStarting ONNX export with onnx %s...' % onnx.__version__)
+        f = opt.weights.replace('.pt', '.onnx')  # filename
+        torch.onnx.export(model, img, f, verbose=False, opset_version=12, input_names=['images'],
+                          output_names=['classes', 'boxes'] if y is None else ['output'],
+                          dynamic_axes={'images': {0: 'batch', 2: 'height', 3: 'width'},  # size(1,3,640,640)
+                                        'output': {0: 'batch', 2: 'y', 3: 'x'}} if opt.dynamic else None)
+
+        # Checks
+        onnx_model = onnx.load(f)  # load onnx model
+        onnx.checker.check_model(onnx_model)  # check onnx model
+        # print(onnx.helper.printable_graph(onnx_model.graph))  # print a human readable model
+        print('ONNX export success, saved as %s' % f)
+    except Exception as e:
+        print('ONNX export failure: %s' % e)
+
+    # CoreML export
+    try:
+        import coremltools as ct
+
+        print('\nStarting CoreML export with coremltools %s...' % ct.__version__)
+        # convert model from torchscript and apply pixel scaling as per detect.py
+        model = ct.convert(ts, inputs=[ct.ImageType(name='image', shape=img.shape, scale=1 / 255.0, bias=[0, 0, 0])])
+        f = opt.weights.replace('.pt', '.mlmodel')  # filename
+        model.save(f)
+        print('CoreML export success, saved as %s' % f)
+    except Exception as e:
+        print('CoreML export failure: %s' % e)
+
+    # Finish
+    print('\nExport complete (%.2fs). Visualize with https://github.com/lutzroeder/netron.' % (time.time() - t))

models/yolo.py

@@ -0,0 +1,843 @@
+import argparse
+import logging
+import sys
+from copy import deepcopy
+
+sys.path.append('./')  # to run '$ python *.py' files in subdirectories
+logger = logging.getLogger(__name__)
+import torch
+from models.common import *
+from models.experimental import *
+from utils.autoanchor import check_anchor_order
+from utils.general import make_divisible, check_file, set_logging
+from utils.torch_utils import time_synchronized, fuse_conv_and_bn, model_info, scale_img, initialize_weights, \
+    select_device, copy_attr
+from utils.loss import SigmoidBin
+
+try:
+    import thop  # for FLOPS computation
+except ImportError:
+    thop = None
+
+
+class Detect(nn.Module):
+    stride = None  # strides computed during build
+    export = False  # onnx export
+    end2end = False
+    include_nms = False
+    concat = False
+
+    def __init__(self, nc=80, anchors=(), ch=()):  # detection layer
+        super(Detect, self).__init__()
+        self.nc = nc  # number of classes
+        self.no = nc + 5  # number of outputs per anchor
+        self.nl = len(anchors)  # number of detection layers
+        self.na = len(anchors[0]) // 2  # number of anchors
+        self.grid = [torch.zeros(1)] * self.nl  # init grid
+        a = torch.tensor(anchors).float().view(self.nl, -1, 2)
+        self.register_buffer('anchors', a)  # shape(nl,na,2)
+        self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2))  # shape(nl,1,na,1,1,2)
+        self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)  # output conv
+
+    def forward(self, x):
+        # x = x.copy()  # for profiling
+        z = []  # inference output
+        self.training |= self.export
+        for i in range(self.nl):
+            x[i] = self.m[i](x[i])  # conv
+            bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
+            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+
+            if not self.training:  # inference
+                if self.grid[i].shape[2:4] != x[i].shape[2:4]:
+                    self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
+                y = x[i].sigmoid()
+                if not torch.onnx.is_in_onnx_export():
+                    y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i]  # xy
+                    y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
+                else:
+                    xy, wh, conf = y.split((2, 2, self.nc + 1), 4)  # y.tensor_split((2, 4, 5), 4)  # torch 1.8.0
+                    xy = xy * (2. * self.stride[i]) + (self.stride[i] * (self.grid[i] - 0.5))  # new xy
+                    wh = wh ** 2 * (4 * self.anchor_grid[i].data)  # new wh
+                    y = torch.cat((xy, wh, conf), 4)
+                z.append(y.view(bs, -1, self.no))
+
+        if self.training:
+            out = x
+        elif self.end2end:
+            out = torch.cat(z, 1)
+        elif self.include_nms:
+            z = self.convert(z)
+            out = (z, )
+        elif self.concat:
+            out = torch.cat(z, 1)
+        else:
+            out = (torch.cat(z, 1), x)
+
+        return out
+
+    @staticmethod
+    def _make_grid(nx=20, ny=20):
+        yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
+        return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
+
+    def convert(self, z):
+        z = torch.cat(z, 1)
+        box = z[:, :, :4]
+        conf = z[:, :, 4:5]
+        score = z[:, :, 5:]
+        score *= conf
+        convert_matrix = torch.tensor([[1, 0, 1, 0], [0, 1, 0, 1], [-0.5, 0, 0.5, 0], [0, -0.5, 0, 0.5]],
+                                           dtype=torch.float32,
+                                           device=z.device)
+        box @= convert_matrix                          
+        return (box, score)
+
+
+class IDetect(nn.Module):
+    stride = None  # strides computed during build
+    export = False  # onnx export
+    end2end = False
+    include_nms = False
+    concat = False
+
+    def __init__(self, nc=80, anchors=(), ch=()):  # detection layer
+        super(IDetect, self).__init__()
+        self.nc = nc  # number of classes
+        self.no = nc + 5  # number of outputs per anchor
+        self.nl = len(anchors)  # number of detection layers
+        self.na = len(anchors[0]) // 2  # number of anchors
+        self.grid = [torch.zeros(1)] * self.nl  # init grid
+        a = torch.tensor(anchors).float().view(self.nl, -1, 2)
+        self.register_buffer('anchors', a)  # shape(nl,na,2)
+        self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2))  # shape(nl,1,na,1,1,2)
+        self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)  # output conv
+        
+        self.ia = nn.ModuleList(ImplicitA(x) for x in ch)
+        self.im = nn.ModuleList(ImplicitM(self.no * self.na) for _ in ch)
+
+    def forward(self, x):
+        # x = x.copy()  # for profiling
+        z = []  # inference output
+        self.training |= self.export
+        for i in range(self.nl):
+            x[i] = self.m[i](self.ia[i](x[i]))  # conv
+            x[i] = self.im[i](x[i])
+            bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
+            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+
+            if not self.training:  # inference
+                if self.grid[i].shape[2:4] != x[i].shape[2:4]:
+                    self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
+
+                y = x[i].sigmoid()
+                y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i]  # xy
+                y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
+                z.append(y.view(bs, -1, self.no))
+
+        return x if self.training else (torch.cat(z, 1), x)
+    
+    def fuseforward(self, x):
+        # x = x.copy()  # for profiling
+        z = []  # inference output
+        self.training |= self.export
+        for i in range(self.nl):
+            x[i] = self.m[i](x[i])  # conv
+            bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
+            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+
+            if not self.training:  # inference
+                if self.grid[i].shape[2:4] != x[i].shape[2:4]:
+                    self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
+
+                y = x[i].sigmoid()
+                if not torch.onnx.is_in_onnx_export():
+                    y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i]  # xy
+                    y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
+                else:
+                    xy, wh, conf = y.split((2, 2, self.nc + 1), 4)  # y.tensor_split((2, 4, 5), 4)  # torch 1.8.0
+                    xy = xy * (2. * self.stride[i]) + (self.stride[i] * (self.grid[i] - 0.5))  # new xy
+                    wh = wh ** 2 * (4 * self.anchor_grid[i].data)  # new wh
+                    y = torch.cat((xy, wh, conf), 4)
+                z.append(y.view(bs, -1, self.no))
+
+        if self.training:
+            out = x
+        elif self.end2end:
+            out = torch.cat(z, 1)
+        elif self.include_nms:
+            z = self.convert(z)
+            out = (z, )
+        elif self.concat:
+            out = torch.cat(z, 1)            
+        else:
+            out = (torch.cat(z, 1), x)
+
+        return out
+    
+    def fuse(self):
+        print("IDetect.fuse")
+        # fuse ImplicitA and Convolution
+        for i in range(len(self.m)):
+            c1,c2,_,_ = self.m[i].weight.shape
+            c1_,c2_, _,_ = self.ia[i].implicit.shape
+            self.m[i].bias += torch.matmul(self.m[i].weight.reshape(c1,c2),self.ia[i].implicit.reshape(c2_,c1_)).squeeze(1)
+
+        # fuse ImplicitM and Convolution
+        for i in range(len(self.m)):
+            c1,c2, _,_ = self.im[i].implicit.shape
+            self.m[i].bias *= self.im[i].implicit.reshape(c2)
+            self.m[i].weight *= self.im[i].implicit.transpose(0,1)
+            
+    @staticmethod
+    def _make_grid(nx=20, ny=20):
+        yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
+        return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
+
+    def convert(self, z):
+        z = torch.cat(z, 1)
+        box = z[:, :, :4]
+        conf = z[:, :, 4:5]
+        score = z[:, :, 5:]
+        score *= conf
+        convert_matrix = torch.tensor([[1, 0, 1, 0], [0, 1, 0, 1], [-0.5, 0, 0.5, 0], [0, -0.5, 0, 0.5]],
+                                           dtype=torch.float32,
+                                           device=z.device)
+        box @= convert_matrix                          
+        return (box, score)
+
+
+class IKeypoint(nn.Module):
+    stride = None  # strides computed during build
+    export = False  # onnx export
+
+    def __init__(self, nc=80, anchors=(), nkpt=17, ch=(), inplace=True, dw_conv_kpt=False):  # detection layer
+        super(IKeypoint, self).__init__()
+        self.nc = nc  # number of classes
+        self.nkpt = nkpt
+        self.dw_conv_kpt = dw_conv_kpt
+        self.no_det=(nc + 5)  # number of outputs per anchor for box and class
+        self.no_kpt = 3*self.nkpt ## number of outputs per anchor for keypoints
+        self.no = self.no_det+self.no_kpt
+        self.nl = len(anchors)  # number of detection layers
+        self.na = len(anchors[0]) // 2  # number of anchors
+        self.grid = [torch.zeros(1)] * self.nl  # init grid
+        self.flip_test = False
+        a = torch.tensor(anchors).float().view(self.nl, -1, 2)
+        self.register_buffer('anchors', a)  # shape(nl,na,2)
+        self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2))  # shape(nl,1,na,1,1,2)
+        self.m = nn.ModuleList(nn.Conv2d(x, self.no_det * self.na, 1) for x in ch)  # output conv
+        
+        self.ia = nn.ModuleList(ImplicitA(x) for x in ch)
+        self.im = nn.ModuleList(ImplicitM(self.no_det * self.na) for _ in ch)
+        
+        if self.nkpt is not None:
+            if self.dw_conv_kpt: #keypoint head is slightly more complex
+                self.m_kpt = nn.ModuleList(
+                            nn.Sequential(DWConv(x, x, k=3), Conv(x,x),
+                                          DWConv(x, x, k=3), Conv(x, x),
+                                          DWConv(x, x, k=3), Conv(x,x),
+                                          DWConv(x, x, k=3), Conv(x, x),
+                                          DWConv(x, x, k=3), Conv(x, x),
+                                          DWConv(x, x, k=3), nn.Conv2d(x, self.no_kpt * self.na, 1)) for x in ch)
+            else: #keypoint head is a single convolution
+                self.m_kpt = nn.ModuleList(nn.Conv2d(x, self.no_kpt * self.na, 1) for x in ch)
+
+        self.inplace = inplace  # use in-place ops (e.g. slice assignment)
+
+    def forward(self, x):
+        # x = x.copy()  # for profiling
+        z = []  # inference output
+        self.training |= self.export
+        for i in range(self.nl):
+            if self.nkpt is None or self.nkpt==0:
+                x[i] = self.im[i](self.m[i](self.ia[i](x[i])))  # conv
+            else :
+                x[i] = torch.cat((self.im[i](self.m[i](self.ia[i](x[i]))), self.m_kpt[i](x[i])), axis=1)
+
+            bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
+            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+            x_det = x[i][..., :6]
+            x_kpt = x[i][..., 6:]
+
+            if not self.training:  # inference
+                if self.grid[i].shape[2:4] != x[i].shape[2:4]:
+                    self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
+                kpt_grid_x = self.grid[i][..., 0:1]
+                kpt_grid_y = self.grid[i][..., 1:2]
+
+                if self.nkpt == 0:
+                    y = x[i].sigmoid()
+                else:
+                    y = x_det.sigmoid()
+
+                if self.inplace:
+                    xy = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i]  # xy
+                    wh = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i].view(1, self.na, 1, 1, 2) # wh
+                    if self.nkpt != 0:
+                        x_kpt[..., 0::3] = (x_kpt[..., ::3] * 2. - 0.5 + kpt_grid_x.repeat(1,1,1,1,17)) * self.stride[i]  # xy
+                        x_kpt[..., 1::3] = (x_kpt[..., 1::3] * 2. - 0.5 + kpt_grid_y.repeat(1,1,1,1,17)) * self.stride[i]  # xy
+                        #x_kpt[..., 0::3] = (x_kpt[..., ::3] + kpt_grid_x.repeat(1,1,1,1,17)) * self.stride[i]  # xy
+                        #x_kpt[..., 1::3] = (x_kpt[..., 1::3] + kpt_grid_y.repeat(1,1,1,1,17)) * self.stride[i]  # xy
+                        #print('=============')
+                        #print(self.anchor_grid[i].shape)
+                        #print(self.anchor_grid[i][...,0].unsqueeze(4).shape)
+                        #print(x_kpt[..., 0::3].shape)
+                        #x_kpt[..., 0::3] = ((x_kpt[..., 0::3].tanh() * 2.) ** 3 * self.anchor_grid[i][...,0].unsqueeze(4).repeat(1,1,1,1,self.nkpt)) + kpt_grid_x.repeat(1,1,1,1,17) * self.stride[i]  # xy
+                        #x_kpt[..., 1::3] = ((x_kpt[..., 1::3].tanh() * 2.) ** 3 * self.anchor_grid[i][...,1].unsqueeze(4).repeat(1,1,1,1,self.nkpt)) + kpt_grid_y.repeat(1,1,1,1,17) * self.stride[i]  # xy
+                        #x_kpt[..., 0::3] = (((x_kpt[..., 0::3].sigmoid() * 4.) ** 2 - 8.) * self.anchor_grid[i][...,0].unsqueeze(4).repeat(1,1,1,1,self.nkpt)) + kpt_grid_x.repeat(1,1,1,1,17) * self.stride[i]  # xy
+                        #x_kpt[..., 1::3] = (((x_kpt[..., 1::3].sigmoid() * 4.) ** 2 - 8.) * self.anchor_grid[i][...,1].unsqueeze(4).repeat(1,1,1,1,self.nkpt)) + kpt_grid_y.repeat(1,1,1,1,17) * self.stride[i]  # xy
+                        x_kpt[..., 2::3] = x_kpt[..., 2::3].sigmoid()
+
+                    y = torch.cat((xy, wh, y[..., 4:], x_kpt), dim = -1)
+
+                else:  # for YOLOv5 on AWS Inferentia https://github.com/ultralytics/yolov5/pull/2953
+                    xy = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i]  # xy
+                    wh = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
+                    if self.nkpt != 0:
+                        y[..., 6:] = (y[..., 6:] * 2. - 0.5 + self.grid[i].repeat((1,1,1,1,self.nkpt))) * self.stride[i]  # xy
+                    y = torch.cat((xy, wh, y[..., 4:]), -1)
+
+                z.append(y.view(bs, -1, self.no))
+
+        return x if self.training else (torch.cat(z, 1), x)
+
+    @staticmethod
+    def _make_grid(nx=20, ny=20):
+        yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
+        return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
+
+
+class IAuxDetect(nn.Module):
+    stride = None  # strides computed during build
+    export = False  # onnx export
+    end2end = False
+    include_nms = False
+    concat = False
+
+    def __init__(self, nc=80, anchors=(), ch=()):  # detection layer
+        super(IAuxDetect, self).__init__()
+        self.nc = nc  # number of classes
+        self.no = nc + 5  # number of outputs per anchor
+        self.nl = len(anchors)  # number of detection layers
+        self.na = len(anchors[0]) // 2  # number of anchors
+        self.grid = [torch.zeros(1)] * self.nl  # init grid
+        a = torch.tensor(anchors).float().view(self.nl, -1, 2)
+        self.register_buffer('anchors', a)  # shape(nl,na,2)
+        self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2))  # shape(nl,1,na,1,1,2)
+        self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch[:self.nl])  # output conv
+        self.m2 = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch[self.nl:])  # output conv
+        
+        self.ia = nn.ModuleList(ImplicitA(x) for x in ch[:self.nl])
+        self.im = nn.ModuleList(ImplicitM(self.no * self.na) for _ in ch[:self.nl])
+
+    def forward(self, x):
+        # x = x.copy()  # for profiling
+        z = []  # inference output
+        self.training |= self.export
+        for i in range(self.nl):
+            x[i] = self.m[i](self.ia[i](x[i]))  # conv
+            x[i] = self.im[i](x[i])
+            bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
+            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+            
+            x[i+self.nl] = self.m2[i](x[i+self.nl])
+            x[i+self.nl] = x[i+self.nl].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+
+            if not self.training:  # inference
+                if self.grid[i].shape[2:4] != x[i].shape[2:4]:
+                    self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
+
+                y = x[i].sigmoid()
+                if not torch.onnx.is_in_onnx_export():
+                    y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i]  # xy
+                    y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
+                else:
+                    xy, wh, conf = y.split((2, 2, self.nc + 1), 4)  # y.tensor_split((2, 4, 5), 4)  # torch 1.8.0
+                    xy = xy * (2. * self.stride[i]) + (self.stride[i] * (self.grid[i] - 0.5))  # new xy
+                    wh = wh ** 2 * (4 * self.anchor_grid[i].data)  # new wh
+                    y = torch.cat((xy, wh, conf), 4)
+                z.append(y.view(bs, -1, self.no))
+
+        return x if self.training else (torch.cat(z, 1), x[:self.nl])
+
+    def fuseforward(self, x):
+        # x = x.copy()  # for profiling
+        z = []  # inference output
+        self.training |= self.export
+        for i in range(self.nl):
+            x[i] = self.m[i](x[i])  # conv
+            bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
+            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+
+            if not self.training:  # inference
+                if self.grid[i].shape[2:4] != x[i].shape[2:4]:
+                    self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
+
+                y = x[i].sigmoid()
+                if not torch.onnx.is_in_onnx_export():
+                    y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i]  # xy
+                    y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
+                else:
+                    xy = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i]  # xy
+                    wh = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i].data  # wh
+                    y = torch.cat((xy, wh, y[..., 4:]), -1)
+                z.append(y.view(bs, -1, self.no))
+
+        if self.training:
+            out = x
+        elif self.end2end:
+            out = torch.cat(z, 1)
+        elif self.include_nms:
+            z = self.convert(z)
+            out = (z, )
+        elif self.concat:
+            out = torch.cat(z, 1)            
+        else:
+            out = (torch.cat(z, 1), x)
+
+        return out
+    
+    def fuse(self):
+        print("IAuxDetect.fuse")
+        # fuse ImplicitA and Convolution
+        for i in range(len(self.m)):
+            c1,c2,_,_ = self.m[i].weight.shape
+            c1_,c2_, _,_ = self.ia[i].implicit.shape
+            self.m[i].bias += torch.matmul(self.m[i].weight.reshape(c1,c2),self.ia[i].implicit.reshape(c2_,c1_)).squeeze(1)
+
+        # fuse ImplicitM and Convolution
+        for i in range(len(self.m)):
+            c1,c2, _,_ = self.im[i].implicit.shape
+            self.m[i].bias *= self.im[i].implicit.reshape(c2)
+            self.m[i].weight *= self.im[i].implicit.transpose(0,1)
+
+    @staticmethod
+    def _make_grid(nx=20, ny=20):
+        yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
+        return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
+
+    def convert(self, z):
+        z = torch.cat(z, 1)
+        box = z[:, :, :4]
+        conf = z[:, :, 4:5]
+        score = z[:, :, 5:]
+        score *= conf
+        convert_matrix = torch.tensor([[1, 0, 1, 0], [0, 1, 0, 1], [-0.5, 0, 0.5, 0], [0, -0.5, 0, 0.5]],
+                                           dtype=torch.float32,
+                                           device=z.device)
+        box @= convert_matrix                          
+        return (box, score)
+
+
+class IBin(nn.Module):
+    stride = None  # strides computed during build
+    export = False  # onnx export
+
+    def __init__(self, nc=80, anchors=(), ch=(), bin_count=21):  # detection layer
+        super(IBin, self).__init__()
+        self.nc = nc  # number of classes
+        self.bin_count = bin_count
+
+        self.w_bin_sigmoid = SigmoidBin(bin_count=self.bin_count, min=0.0, max=4.0)
+        self.h_bin_sigmoid = SigmoidBin(bin_count=self.bin_count, min=0.0, max=4.0)
+        # classes, x,y,obj
+        self.no = nc + 3 + \
+            self.w_bin_sigmoid.get_length() + self.h_bin_sigmoid.get_length()   # w-bce, h-bce
+            # + self.x_bin_sigmoid.get_length() + self.y_bin_sigmoid.get_length()
+        
+        self.nl = len(anchors)  # number of detection layers
+        self.na = len(anchors[0]) // 2  # number of anchors
+        self.grid = [torch.zeros(1)] * self.nl  # init grid
+        a = torch.tensor(anchors).float().view(self.nl, -1, 2)
+        self.register_buffer('anchors', a)  # shape(nl,na,2)
+        self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2))  # shape(nl,1,na,1,1,2)
+        self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)  # output conv
+        
+        self.ia = nn.ModuleList(ImplicitA(x) for x in ch)
+        self.im = nn.ModuleList(ImplicitM(self.no * self.na) for _ in ch)
+
+    def forward(self, x):
+
+        #self.x_bin_sigmoid.use_fw_regression = True
+        #self.y_bin_sigmoid.use_fw_regression = True
+        self.w_bin_sigmoid.use_fw_regression = True
+        self.h_bin_sigmoid.use_fw_regression = True
+        
+        # x = x.copy()  # for profiling
+        z = []  # inference output
+        self.training |= self.export
+        for i in range(self.nl):
+            x[i] = self.m[i](self.ia[i](x[i]))  # conv
+            x[i] = self.im[i](x[i])
+            bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
+            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+
+            if not self.training:  # inference
+                if self.grid[i].shape[2:4] != x[i].shape[2:4]:
+                    self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
+
+                y = x[i].sigmoid()
+                y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i]  # xy
+                #y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
+                
+
+                #px = (self.x_bin_sigmoid.forward(y[..., 0:12]) + self.grid[i][..., 0]) * self.stride[i]
+                #py = (self.y_bin_sigmoid.forward(y[..., 12:24]) + self.grid[i][..., 1]) * self.stride[i]
+
+                pw = self.w_bin_sigmoid.forward(y[..., 2:24]) * self.anchor_grid[i][..., 0]
+                ph = self.h_bin_sigmoid.forward(y[..., 24:46]) * self.anchor_grid[i][..., 1]
+
+                #y[..., 0] = px
+                #y[..., 1] = py
+                y[..., 2] = pw
+                y[..., 3] = ph
+                
+                y = torch.cat((y[..., 0:4], y[..., 46:]), dim=-1)
+                
+                z.append(y.view(bs, -1, y.shape[-1]))
+
+        return x if self.training else (torch.cat(z, 1), x)
+
+    @staticmethod
+    def _make_grid(nx=20, ny=20):
+        yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
+        return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
+
+
+class Model(nn.Module):
+    def __init__(self, cfg='yolor-csp-c.yaml', ch=3, nc=None, anchors=None):  # model, input channels, number of classes
+        super(Model, self).__init__()
+        self.traced = False
+        if isinstance(cfg, dict):
+            self.yaml = cfg  # model dict
+        else:  # is *.yaml
+            import yaml  # for torch hub
+            self.yaml_file = Path(cfg).name
+            with open(cfg) as f:
+                self.yaml = yaml.load(f, Loader=yaml.SafeLoader)  # model dict
+
+        # Define model
+        ch = self.yaml['ch'] = self.yaml.get('ch', ch)  # input channels
+        if nc and nc != self.yaml['nc']:
+            logger.info(f"Overriding model.yaml nc={self.yaml['nc']} with nc={nc}")
+            self.yaml['nc'] = nc  # override yaml value
+        if anchors:
+            logger.info(f'Overriding model.yaml anchors with anchors={anchors}')
+            self.yaml['anchors'] = round(anchors)  # override yaml value
+        self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch])  # model, savelist
+        self.names = [str(i) for i in range(self.yaml['nc'])]  # default names
+        # print([x.shape for x in self.forward(torch.zeros(1, ch, 64, 64))])
+
+        # Build strides, anchors
+        m = self.model[-1]  # Detect()
+        if isinstance(m, Detect):
+            s = 256  # 2x min stride
+            m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))])  # forward
+            check_anchor_order(m)
+            m.anchors /= m.stride.view(-1, 1, 1)
+            self.stride = m.stride
+            self._initialize_biases()  # only run once
+            # print('Strides: %s' % m.stride.tolist())
+        if isinstance(m, IDetect):
+            s = 256  # 2x min stride
+            m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))])  # forward
+            check_anchor_order(m)
+            m.anchors /= m.stride.view(-1, 1, 1)
+            self.stride = m.stride
+            self._initialize_biases()  # only run once
+            # print('Strides: %s' % m.stride.tolist())
+        if isinstance(m, IAuxDetect):
+            s = 256  # 2x min stride
+            m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))[:4]])  # forward
+            #print(m.stride)
+            check_anchor_order(m)
+            m.anchors /= m.stride.view(-1, 1, 1)
+            self.stride = m.stride
+            self._initialize_aux_biases()  # only run once
+            # print('Strides: %s' % m.stride.tolist())
+        if isinstance(m, IBin):
+            s = 256  # 2x min stride
+            m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))])  # forward
+            check_anchor_order(m)
+            m.anchors /= m.stride.view(-1, 1, 1)
+            self.stride = m.stride
+            self._initialize_biases_bin()  # only run once
+            # print('Strides: %s' % m.stride.tolist())
+        if isinstance(m, IKeypoint):
+            s = 256  # 2x min stride
+            m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))])  # forward
+            check_anchor_order(m)
+            m.anchors /= m.stride.view(-1, 1, 1)
+            self.stride = m.stride
+            self._initialize_biases_kpt()  # only run once
+            # print('Strides: %s' % m.stride.tolist())
+
+        # Init weights, biases
+        initialize_weights(self)
+        self.info()
+        logger.info('')
+
+    def forward(self, x, augment=False, profile=False):
+        if augment:
+            img_size = x.shape[-2:]  # height, width
+            s = [1, 0.83, 0.67]  # scales
+            f = [None, 3, None]  # flips (2-ud, 3-lr)
+            y = []  # outputs
+            for si, fi in zip(s, f):
+                xi = scale_img(x.flip(fi) if fi else x, si, gs=int(self.stride.max()))
+                yi = self.forward_once(xi)[0]  # forward
+                # cv2.imwrite(f'img_{si}.jpg', 255 * xi[0].cpu().numpy().transpose((1, 2, 0))[:, :, ::-1])  # save
+                yi[..., :4] /= si  # de-scale
+                if fi == 2:
+                    yi[..., 1] = img_size[0] - yi[..., 1]  # de-flip ud
+                elif fi == 3:
+                    yi[..., 0] = img_size[1] - yi[..., 0]  # de-flip lr
+                y.append(yi)
+            return torch.cat(y, 1), None  # augmented inference, train
+        else:
+            return self.forward_once(x, profile)  # single-scale inference, train
+
+    def forward_once(self, x, profile=False):
+        y, dt = [], []  # outputs
+        for m in self.model:
+            if m.f != -1:  # if not from previous layer
+                x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f]  # from earlier layers
+
+            if not hasattr(self, 'traced'):
+                self.traced=False
+
+            if self.traced:
+                if isinstance(m, Detect) or isinstance(m, IDetect) or isinstance(m, IAuxDetect) or isinstance(m, IKeypoint):
+                    break
+
+            if profile:
+                c = isinstance(m, (Detect, IDetect, IAuxDetect, IBin))
+                o = thop.profile(m, inputs=(x.copy() if c else x,), verbose=False)[0] / 1E9 * 2 if thop else 0  # FLOPS
+                for _ in range(10):
+                    m(x.copy() if c else x)
+                t = time_synchronized()
+                for _ in range(10):
+                    m(x.copy() if c else x)
+                dt.append((time_synchronized() - t) * 100)
+                print('%10.1f%10.0f%10.1fms %-40s' % (o, m.np, dt[-1], m.type))
+
+            x = m(x)  # run
+            
+            y.append(x if m.i in self.save else None)  # save output
+
+        if profile:
+            print('%.1fms total' % sum(dt))
+        return x
+
+    def _initialize_biases(self, cf=None):  # initialize biases into Detect(), cf is class frequency
+        # https://arxiv.org/abs/1708.02002 section 3.3
+        # cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
+        m = self.model[-1]  # Detect() module
+        for mi, s in zip(m.m, m.stride):  # from
+            b = mi.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
+            b.data[:, 4] += math.log(8 / (640 / s) ** 2)  # obj (8 objects per 640 image)
+            b.data[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum())  # cls
+            mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
+
+    def _initialize_aux_biases(self, cf=None):  # initialize biases into Detect(), cf is class frequency
+        # https://arxiv.org/abs/1708.02002 section 3.3
+        # cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
+        m = self.model[-1]  # Detect() module
+        for mi, mi2, s in zip(m.m, m.m2, m.stride):  # from
+            b = mi.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
+            b.data[:, 4] += math.log(8 / (640 / s) ** 2)  # obj (8 objects per 640 image)
+            b.data[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum())  # cls
+            mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
+            b2 = mi2.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
+            b2.data[:, 4] += math.log(8 / (640 / s) ** 2)  # obj (8 objects per 640 image)
+            b2.data[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum())  # cls
+            mi2.bias = torch.nn.Parameter(b2.view(-1), requires_grad=True)
+
+    def _initialize_biases_bin(self, cf=None):  # initialize biases into Detect(), cf is class frequency
+        # https://arxiv.org/abs/1708.02002 section 3.3
+        # cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
+        m = self.model[-1]  # Bin() module
+        bc = m.bin_count
+        for mi, s in zip(m.m, m.stride):  # from
+            b = mi.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
+            old = b[:, (0,1,2,bc+3)].data
+            obj_idx = 2*bc+4
+            b[:, :obj_idx].data += math.log(0.6 / (bc + 1 - 0.99))
+            b[:, obj_idx].data += math.log(8 / (640 / s) ** 2)  # obj (8 objects per 640 image)
+            b[:, (obj_idx+1):].data += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum())  # cls
+            b[:, (0,1,2,bc+3)].data = old
+            mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
+
+    def _initialize_biases_kpt(self, cf=None):  # initialize biases into Detect(), cf is class frequency
+        # https://arxiv.org/abs/1708.02002 section 3.3
+        # cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
+        m = self.model[-1]  # Detect() module
+        for mi, s in zip(m.m, m.stride):  # from
+            b = mi.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
+            b.data[:, 4] += math.log(8 / (640 / s) ** 2)  # obj (8 objects per 640 image)
+            b.data[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum())  # cls
+            mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
+
+    def _print_biases(self):
+        m = self.model[-1]  # Detect() module
+        for mi in m.m:  # from
+            b = mi.bias.detach().view(m.na, -1).T  # conv.bias(255) to (3,85)
+            print(('%6g Conv2d.bias:' + '%10.3g' * 6) % (mi.weight.shape[1], *b[:5].mean(1).tolist(), b[5:].mean()))
+
+    # def _print_weights(self):
+    #     for m in self.model.modules():
+    #         if type(m) is Bottleneck:
+    #             print('%10.3g' % (m.w.detach().sigmoid() * 2))  # shortcut weights
+
+    def fuse(self):  # fuse model Conv2d() + BatchNorm2d() layers
+        print('Fusing layers... ')
+        for m in self.model.modules():
+            if isinstance(m, RepConv):
+                #print(f" fuse_repvgg_block")
+                m.fuse_repvgg_block()
+            elif isinstance(m, RepConv_OREPA):
+                #print(f" switch_to_deploy")
+                m.switch_to_deploy()
+            elif type(m) is Conv and hasattr(m, 'bn'):
+                m.conv = fuse_conv_and_bn(m.conv, m.bn)  # update conv
+                delattr(m, 'bn')  # remove batchnorm
+                m.forward = m.fuseforward  # update forward
+            elif isinstance(m, (IDetect, IAuxDetect)):
+                m.fuse()
+                m.forward = m.fuseforward
+        self.info()
+        return self
+
+    def nms(self, mode=True):  # add or remove NMS module
+        present = type(self.model[-1]) is NMS  # last layer is NMS
+        if mode and not present:
+            print('Adding NMS... ')
+            m = NMS()  # module
+            m.f = -1  # from
+            m.i = self.model[-1].i + 1  # index
+            self.model.add_module(name='%s' % m.i, module=m)  # add
+            self.eval()
+        elif not mode and present:
+            print('Removing NMS... ')
+            self.model = self.model[:-1]  # remove
+        return self
+
+    def autoshape(self):  # add autoShape module
+        print('Adding autoShape... ')
+        m = autoShape(self)  # wrap model
+        copy_attr(m, self, include=('yaml', 'nc', 'hyp', 'names', 'stride'), exclude=())  # copy attributes
+        return m
+
+    def info(self, verbose=False, img_size=640):  # print model information
+        model_info(self, verbose, img_size)
+
+
+def parse_model(d, ch):  # model_dict, input_channels(3)
+    logger.info('\n%3s%18s%3s%10s  %-40s%-30s' % ('', 'from', 'n', 'params', 'module', 'arguments'))
+    anchors, nc, gd, gw = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple']
+    na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors  # number of anchors
+    no = na * (nc + 5)  # number of outputs = anchors * (classes + 5)
+
+    layers, save, c2 = [], [], ch[-1]  # layers, savelist, ch out
+    for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']):  # from, number, module, args
+        m = eval(m) if isinstance(m, str) else m  # eval strings
+        for j, a in enumerate(args):
+            try:
+                args[j] = eval(a) if isinstance(a, str) else a  # eval strings
+            except:
+                pass
+
+        n = max(round(n * gd), 1) if n > 1 else n  # depth gain
+        if m in [nn.Conv2d, Conv, RobustConv, RobustConv2, DWConv, GhostConv, RepConv, RepConv_OREPA, DownC, 
+                 SPP, SPPF, SPPCSPC, GhostSPPCSPC, MixConv2d, Focus, Stem, GhostStem, CrossConv, 
+                 Bottleneck, BottleneckCSPA, BottleneckCSPB, BottleneckCSPC, 
+                 RepBottleneck, RepBottleneckCSPA, RepBottleneckCSPB, RepBottleneckCSPC,  
+                 Res, ResCSPA, ResCSPB, ResCSPC, 
+                 RepRes, RepResCSPA, RepResCSPB, RepResCSPC, 
+                 ResX, ResXCSPA, ResXCSPB, ResXCSPC, 
+                 RepResX, RepResXCSPA, RepResXCSPB, RepResXCSPC, 
+                 Ghost, GhostCSPA, GhostCSPB, GhostCSPC,
+                 SwinTransformerBlock, STCSPA, STCSPB, STCSPC,
+                 SwinTransformer2Block, ST2CSPA, ST2CSPB, ST2CSPC]:
+            c1, c2 = ch[f], args[0]
+            if c2 != no:  # if not output
+                c2 = make_divisible(c2 * gw, 8)
+
+            args = [c1, c2, *args[1:]]
+            if m in [DownC, SPPCSPC, GhostSPPCSPC, 
+                     BottleneckCSPA, BottleneckCSPB, BottleneckCSPC, 
+                     RepBottleneckCSPA, RepBottleneckCSPB, RepBottleneckCSPC, 
+                     ResCSPA, ResCSPB, ResCSPC, 
+                     RepResCSPA, RepResCSPB, RepResCSPC, 
+                     ResXCSPA, ResXCSPB, ResXCSPC, 
+                     RepResXCSPA, RepResXCSPB, RepResXCSPC,
+                     GhostCSPA, GhostCSPB, GhostCSPC,
+                     STCSPA, STCSPB, STCSPC,
+                     ST2CSPA, ST2CSPB, ST2CSPC]:
+                args.insert(2, n)  # number of repeats
+                n = 1
+        elif m is nn.BatchNorm2d:
+            args = [ch[f]]
+        elif m is Concat:
+            c2 = sum([ch[x] for x in f])
+        elif m is Chuncat:
+            c2 = sum([ch[x] for x in f])
+        elif m is Shortcut:
+            c2 = ch[f[0]]
+        elif m is Foldcut:
+            c2 = ch[f] // 2
+        elif m in [Detect, IDetect, IAuxDetect, IBin, IKeypoint]:
+            args.append([ch[x] for x in f])
+            if isinstance(args[1], int):  # number of anchors
+                args[1] = [list(range(args[1] * 2))] * len(f)
+        elif m is ReOrg:
+            c2 = ch[f] * 4
+        elif m is Contract:
+            c2 = ch[f] * args[0] ** 2
+        elif m is Expand:
+            c2 = ch[f] // args[0] ** 2
+        else:
+            c2 = ch[f]
+
+        m_ = nn.Sequential(*[m(*args) for _ in range(n)]) if n > 1 else m(*args)  # module
+        t = str(m)[8:-2].replace('__main__.', '')  # module type
+        np = sum([x.numel() for x in m_.parameters()])  # number params
+        m_.i, m_.f, m_.type, m_.np = i, f, t, np  # attach index, 'from' index, type, number params
+        logger.info('%3s%18s%3s%10.0f  %-40s%-30s' % (i, f, n, np, t, args))  # print
+        save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1)  # append to savelist
+        layers.append(m_)
+        if i == 0:
+            ch = []
+        ch.append(c2)
+    return nn.Sequential(*layers), sorted(save)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--cfg', type=str, default='yolor-csp-c.yaml', help='model.yaml')
+    parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
+    parser.add_argument('--profile', action='store_true', help='profile model speed')
+    opt = parser.parse_args()
+    opt.cfg = check_file(opt.cfg)  # check file
+    set_logging()
+    device = select_device(opt.device)
+
+    # Create model
+    model = Model(opt.cfg).to(device)
+    model.train()
+    
+    if opt.profile:
+        img = torch.rand(1, 3, 640, 640).to(device)
+        y = model(img, profile=True)
+
+    # Profile
+    # img = torch.rand(8 if torch.cuda.is_available() else 1, 3, 640, 640).to(device)
+    # y = model(img, profile=True)
+
+    # Tensorboard
+    # from torch.utils.tensorboard import SummaryWriter
+    # tb_writer = SummaryWriter()
+    # print("Run 'tensorboard --logdir=models/runs' to view tensorboard at http://localhost:6006/")
+    # tb_writer.add_graph(model.model, img)  # add model to tensorboard
+    # tb_writer.add_image('test', img[0], dataformats='CWH')  # add model to tensorboard