remove useless codes

models/yolov8/yolov8_head.py

@@ -124,154 +124,3 @@ class Yolov8DetHead(nn.Module):
             reg_feats.append(reg_feat)
 
         return cls_feats, reg_feats
-
-
-# -------------------- Segmentation Head --------------------
-## Single-level Segmentation Head (not complete yet)
-class SegHead(nn.Module):
-    def __init__(self,
-                 in_dim       :int  = 256,
-                 cls_head_dim :int  = 256,
-                 reg_head_dim :int  = 256,
-                 seg_head_dim :int  = 256,
-                 num_cls_head :int  = 2,
-                 num_reg_head :int  = 2,
-                 num_seg_head :int  = 2,
-                 act_type     :str  = "silu",
-                 norm_type    :str  = "BN",
-                 depthwise    :bool = False):
-        super().__init__()
-        # --------- Basic Parameters ----------
-        self.in_dim = in_dim
-        self.num_cls_head = num_cls_head
-        self.num_reg_head = num_reg_head
-        self.num_seg_head = num_seg_head
-        self.act_type = act_type
-        self.norm_type = norm_type
-        self.depthwise = depthwise
-        
-        # --------- Network Parameters ----------
-        ## cls head
-        cls_feats = []
-        self.cls_head_dim = cls_head_dim
-        for i in range(num_cls_head):
-            if i == 0:
-                cls_feats.append(
-                    BasicConv(in_dim, self.cls_head_dim,
-                              kernel_size=3, padding=1, stride=1, 
-                              act_type=act_type,
-                              norm_type=norm_type,
-                              depthwise=depthwise)
-                              )
-            else:
-                cls_feats.append(
-                    BasicConv(self.cls_head_dim, self.cls_head_dim,
-                              kernel_size=3, padding=1, stride=1, 
-                              act_type=act_type,
-                              norm_type=norm_type,
-                              depthwise=depthwise)
-                              )
-        ## reg head
-        reg_feats = []
-        self.reg_head_dim = reg_head_dim
-        for i in range(num_reg_head):
-            if i == 0:
-                reg_feats.append(
-                    BasicConv(in_dim, self.reg_head_dim,
-                              kernel_size=3, padding=1, stride=1, 
-                              act_type=act_type,
-                              norm_type=norm_type,
-                              depthwise=depthwise)
-                              )
-            else:
-                reg_feats.append(
-                    BasicConv(self.reg_head_dim, self.reg_head_dim,
-                              kernel_size=3, padding=1, stride=1, 
-                              act_type=act_type,
-                              norm_type=norm_type,
-                              depthwise=depthwise)
-                              )
-        ## seg head
-        seg_feats = []
-        self.seg_head_dim = seg_head_dim
-        for i in range(num_reg_head):
-            if i == 0:
-                seg_feats.append(
-                    BasicConv(in_dim, self.seg_head_dim,
-                              kernel_size=3, padding=1, stride=1, 
-                              act_type=act_type,
-                              norm_type=norm_type,
-                              depthwise=depthwise)
-                              )
-            else:
-                seg_feats.append(
-                    BasicConv(self.seg_head_dim, self.seg_head_dim,
-                              kernel_size=3, padding=1, stride=1, 
-                              act_type=act_type,
-                              norm_type=norm_type,
-                              depthwise=depthwise)
-                              )
-        self.cls_feats = nn.Sequential(*cls_feats)
-        self.reg_feats = nn.Sequential(*reg_feats)
-        self.seg_feats = nn.Sequential(*seg_feats)
-
-        self.init_weights()
-        
-    def init_weights(self):
-        """Initialize the parameters."""
-        for m in self.modules():
-            if isinstance(m, torch.nn.Conv2d):
-                # In order to be consistent with the source code,
-                # reset the Conv2d initialization parameters
-                m.reset_parameters()
-
-    def forward(self, x):
-        """
-            in_feats: (Tensor) [B, C, H, W]
-        """
-        cls_feats = self.cls_feats(x)
-        reg_feats = self.reg_feats(x)
-        seg_feats = self.seg_feats(x)
-
-        return cls_feats, reg_feats, seg_feats
-
-## Multi-level Segmentation Head (not complete yet)
-class YoloSegHead(nn.Module):
-    def __init__(self, cfg, in_dims):
-        super().__init__()
-        ## ----------- Network Parameters -----------
-        self.multi_level_heads = nn.ModuleList(
-            [SegHead(in_dim       = in_dims[level],
-                     cls_head_dim = max(in_dims[0], min(cfg.num_classes, 100)),
-                     reg_head_dim = max(in_dims[0]//4, 16, 4*cfg.reg_max),
-                     seg_head_dim = in_dims[0],
-                     num_cls_head = cfg.num_cls_head,
-                     num_reg_head = cfg.num_reg_head,
-                     num_seg_head = cfg.num_seg_head,
-                     act_type     = cfg.head_act,
-                     norm_type    = cfg.head_norm,
-                     depthwise    = cfg.head_depthwise)
-                     for level in range(cfg.num_levels)
-                     ])
-        # --------- Basic Parameters ----------
-        self.in_dims = in_dims
-        self.cls_head_dim = self.multi_level_heads[0].cls_head_dim
-        self.reg_head_dim = self.multi_level_heads[0].reg_head_dim
-        self.seg_head_dim = self.multi_level_heads[0].seg_head_dim
-
-    def forward(self, feats):
-        """
-            feats: List[(Tensor)] [[B, C, H, W], ...]
-        """
-        cls_feats = []
-        reg_feats = []
-        seg_feats = []
-        for feat, head in zip(feats, self.multi_level_heads):
-            # ---------------- Pred ----------------
-            cls_feat, reg_feat, seg_feat = head(feat)
-
-            cls_feats.append(cls_feat)
-            reg_feats.append(reg_feat)
-            seg_feats.append(seg_feat)
-
-        return cls_feats, reg_feats, seg_feats

models/yolov8/yolov8_pred.py

@@ -151,165 +151,3 @@ class Yolov8DetPredLayer(nn.Module):
                    }
 
         return outputs
-
-
-# -------------------- Segmentation Pred Layer --------------------
-## Single-level pred layer (not complete yet)
-class SegPredLayer(nn.Module):
-    def __init__(self,
-                 cls_dim     :int = 256,
-                 reg_dim     :int = 256,
-                 seg_dim     :int = 256,
-                 stride      :int = 32,
-                 reg_max     :int = 16,
-                 num_classes :int = 80,
-                 num_coords  :int = 4):
-        super().__init__()
-        # --------- Basic Parameters ----------
-        self.stride = stride
-        self.cls_dim = cls_dim
-        self.reg_dim = reg_dim
-        self.seg_dim = seg_dim
-        self.reg_max = reg_max
-        self.num_classes = num_classes
-        self.num_coords = num_coords
-
-        # --------- Network Parameters ----------
-        self.cls_pred = nn.Conv2d(cls_dim, num_classes, kernel_size=1)
-        self.reg_pred = nn.Conv2d(reg_dim, num_coords, kernel_size=1)                
-        self.seg_pred = nn.Conv2d(seg_dim, 1, kernel_size=1)                
-
-        self.init_bias()
-        
-    def init_bias(self):
-        # cls pred bias
-        b = self.cls_pred.bias.view(1, -1)
-        b.data.fill_(math.log(5 / self.num_classes / (640. / self.stride) ** 2))
-        self.cls_pred.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
-        # reg pred bias
-        b = self.reg_pred.bias.view(-1, )
-        b.data.fill_(1.0)
-        self.reg_pred.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
-        # seg pred bias
-        b = self.seg_pred.bias.view(-1, )
-        b.data.fill_(1.0)
-        self.seg_pred.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
-
-    def generate_anchors(self, fmp_size):
-        """
-            fmp_size: (List) [H, W]
-        """
-        # generate grid cells
-        fmp_h, fmp_w = fmp_size
-        anchor_y, anchor_x = torch.meshgrid([torch.arange(fmp_h), torch.arange(fmp_w)])
-        # [H, W, 2] -> [HW, 2]
-        anchors = torch.stack([anchor_x, anchor_y], dim=-1).float().view(-1, 2)
-        anchors += 0.5  # add center offset
-        anchors *= self.stride
-
-        return anchors
-        
-    def forward(self, cls_feat, reg_feat, seg_feat):
-        # pred
-        cls_pred = self.cls_pred(cls_feat)
-        reg_pred = self.reg_pred(reg_feat)
-        seg_pred = self.seg_pred(seg_feat)
-
-        # generate anchor boxes: [M, 4]
-        B, _, H, W = cls_pred.size()
-        fmp_size = [H, W]
-        anchors = self.generate_anchors(fmp_size)
-        anchors = anchors.to(cls_pred.device)
-        # stride tensor: [M, 1]
-        stride_tensor = torch.ones_like(anchors[..., :1]) * self.stride
-        
-        # [B, C, H, W] -> [B, H, W, C] -> [B, M, C]
-        cls_pred = cls_pred.permute(0, 2, 3, 1).contiguous().view(B, -1, self.num_classes)
-        reg_pred = reg_pred.permute(0, 2, 3, 1).contiguous().view(B, -1, 4*self.reg_max)
-        
-        # output dict
-        outputs = {"pred_cls": cls_pred,            # List(Tensor) [B, M, C]
-                   "pred_reg": reg_pred,            # List(Tensor) [B, M, 4*(reg_max)]
-                   "anchors": anchors,              # List(Tensor) [M, 2]
-                   "strides": self.stride,          # List(Int) = [8, 16, 32]
-                   "stride_tensor": stride_tensor   # List(Tensor) [M, 1]
-                   }
-
-        return outputs
-
-## Multi-level pred layer
-class YoloSegPredLayer(nn.Module):
-    def __init__(self,
-                 cfg,
-                 cls_dim,
-                 reg_dim,
-                 seg_dim,
-                 ):
-        super().__init__()
-        # --------- Basic Parameters ----------
-        self.cfg = cfg
-        self.cls_dim = cls_dim
-        self.reg_dim = reg_dim
-        self.seg_dim = seg_dim
-
-        # ----------- Network Parameters -----------
-        ## pred layers
-        self.multi_level_preds = nn.ModuleList(
-            [SegPredLayer(cls_dim     = cls_dim,
-                          reg_dim     = reg_dim,
-                          seg_dim     = seg_dim,
-                          stride      = cfg.out_stride[level],
-                          reg_max     = cfg.reg_max,
-                          num_classes = cfg.num_classes,
-                          num_coords  = 4 * cfg.reg_max)
-                          for level in range(cfg.num_levels)
-                          ])
-        ## proj conv
-        proj_init = torch.arange(cfg.reg_max, dtype=torch.float)
-        self.proj_conv = nn.Conv2d(cfg.reg_max, 1, kernel_size=1, bias=False).requires_grad_(False)
-        self.proj_conv.weight.data[:] = nn.Parameter(proj_init.view([1, cfg.reg_max, 1, 1]), requires_grad=False)
-
-    def forward(self, cls_feats, reg_feats, seg_feats):
-        all_anchors = []
-        all_strides = []
-        all_cls_preds = []
-        all_reg_preds = []
-        all_seg_preds = []
-        all_box_preds = []
-        for level in range(self.cfg.num_levels):
-            # -------------- Single-level prediction --------------
-            outputs = self.multi_level_preds[level](cls_feats[level], reg_feats[level], seg_feats[level])
-
-            # -------------- Decode bbox --------------
-            B, M = outputs["pred_reg"].shape[:2]
-            # [B, M, 4*(reg_max)] -> [B, M, 4, reg_max]
-            delta_pred = outputs["pred_reg"].reshape([B, M, 4, self.cfg.reg_max])
-            # [B, M, 4, reg_max] -> [B, reg_max, 4, M]
-            delta_pred = delta_pred.permute(0, 3, 2, 1).contiguous()
-            # [B, reg_max, 4, M] -> [B, 1, 4, M]
-            delta_pred = self.proj_conv(F.softmax(delta_pred, dim=1))
-            # [B, 1, 4, M] -> [B, 4, M] -> [B, M, 4]
-            delta_pred = delta_pred.view(B, 4, M).permute(0, 2, 1).contiguous()
-            ## tlbr -> xyxy
-            x1y1_pred = outputs["anchors"][None] - delta_pred[..., :2] * self.cfg.out_stride[level]
-            x2y2_pred = outputs["anchors"][None] + delta_pred[..., 2:] * self.cfg.out_stride[level]
-            box_pred = torch.cat([x1y1_pred, x2y2_pred], dim=-1)
-
-            # collect results
-            all_cls_preds.append(outputs["pred_cls"])
-            all_reg_preds.append(outputs["pred_reg"])
-            all_box_preds.append(box_pred)
-            all_anchors.append(outputs["anchors"])
-            all_strides.append(outputs["stride_tensor"])
-        
-        # output dict
-        outputs = {"pred_cls":      all_cls_preds,         # List(Tensor) [B, M, C]
-                   "pred_reg":      all_reg_preds,         # List(Tensor) [B, M, 4*(reg_max)]
-                   "pred_box":      all_box_preds,         # List(Tensor) [B, M, 4]
-                   "anchors":       all_anchors,           # List(Tensor) [M, 2]
-                   "stride_tensor": all_strides,           # List(Tensor) [M, 1]
-                   "strides":       self.cfg.out_stride,   # List(Int) = [8, 16, 32]
-                   }
-
-        return outputs
-