Refactor `Detect()` anchors for ONNX <> OpenCV DNN compatibility (#4833)

* refactor anchors and anchor_grid in Detect Layer

* fix CI failures by adding compatibility

* fix tf failure

* fix different devices errors

* Cleanup

* fix anchors overwriting issue

* better refactoring

* Remove self.anchor_grid shape check (redundant with self.grid check)

Also PEP8 / 120 line width

* Convert _make_grid() from static to dynamic method

* Remove anchor_grid.to(device)

clone() should already clone to same device as self.anchors

* fix different devices error

Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>

models/common.py

@@ -295,6 +295,8 @@ class AutoShape(nn.Module):
         m = self.model.model[-1]  # Detect()
         m.stride = fn(m.stride)
         m.grid = list(map(fn, m.grid))
+        if isinstance(m.anchor_grid, list):
+            m.anchor_grid = list(map(fn, m.anchor_grid))
         return self
 
     @torch.no_grad()

models/experimental.py

@@ -102,6 +102,10 @@ def attempt_load(weights, map_location=None, inplace=True, fuse=True):
     for m in model.modules():
         if type(m) in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU, Detect, Model]:
             m.inplace = inplace  # pytorch 1.7.0 compatibility
+            if type(m) is Detect:
+                if not isinstance(m.anchor_grid, list):  # new Detect Layer compatibility
+                    delattr(m, 'anchor_grid')
+                    setattr(m, 'anchor_grid', [torch.zeros(1)] * m.nl)
         elif type(m) is Conv:
             m._non_persistent_buffers_set = set()  # pytorch 1.6.0 compatibility
 

models/tf.py

@@ -193,7 +193,7 @@ class TFDetect(keras.layers.Layer):
         self.na = len(anchors[0]) // 2  # number of anchors
         self.grid = [tf.zeros(1)] * self.nl  # init grid
         self.anchors = tf.convert_to_tensor(w.anchors.numpy(), dtype=tf.float32)
-        self.anchor_grid = tf.reshape(tf.convert_to_tensor(w.anchor_grid.numpy(), dtype=tf.float32),
+        self.anchor_grid = tf.reshape(self.anchors * tf.reshape(self.stride, [self.nl, 1, 1]),
                                       [self.nl, 1, -1, 1, 2])
         self.m = [TFConv2d(x, self.no * self.na, 1, w=w.m[i]) for i, x in enumerate(ch)]
         self.training = False  # set to False after building model

models/yolo.py

@@ -44,9 +44,8 @@ class Detect(nn.Module):
         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.anchor_grid = [torch.zeros(1)] * self.nl  # init anchor grid
+        self.register_buffer('anchors', torch.tensor(anchors).float().view(self.nl, -1, 2))  # shape(nl,na,2)
         self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)  # output conv
         self.inplace = inplace  # use in-place ops (e.g. slice assignment)
 
@@ -59,7 +58,7 @@ class Detect(nn.Module):
 
             if not self.training:  # inference
                 if self.grid[i].shape[2:4] != x[i].shape[2:4] or self.onnx_dynamic:
-                    self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
+                    self.grid[i], self.anchor_grid[i] = self._make_grid(nx, ny, i)
 
                 y = x[i].sigmoid()
                 if self.inplace:
@@ -67,16 +66,19 @@ class Detect(nn.Module):
                     y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
                 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].view(1, self.na, 1, 1, 2)  # wh
+                    wh = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
                     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()
+    def _make_grid(self, nx=20, ny=20, i=0):
+        d = self.anchors[i].device
+        yv, xv = torch.meshgrid([torch.arange(ny).to(d), torch.arange(nx).to(d)])
+        grid = torch.stack((xv, yv), 2).expand((1, self.na, ny, nx, 2)).float()
+        anchor_grid = (self.anchors[i].clone() * self.stride[i]) \
+            .view((1, self.na, 1, 1, 2)).expand((1, self.na, ny, nx, 2)).float()
+        return grid, anchor_grid
 
 
 class Model(nn.Module):
@@ -239,6 +241,8 @@ class Model(nn.Module):
         if isinstance(m, Detect):
             m.stride = fn(m.stride)
             m.grid = list(map(fn, m.grid))
+            if isinstance(m.anchor_grid, list):
+                m.anchor_grid = list(map(fn, m.anchor_grid))
         return self
 
 

utils/autoanchor.py

@@ -15,13 +15,12 @@ from utils.general import colorstr
 
 def check_anchor_order(m):
     # Check anchor order against stride order for YOLOv5 Detect() module m, and correct if necessary
-    a = m.anchor_grid.prod(-1).view(-1)  # anchor area
+    a = m.anchors.prod(-1).view(-1)  # anchor area
     da = a[-1] - a[0]  # delta a
     ds = m.stride[-1] - m.stride[0]  # delta s
     if da.sign() != ds.sign():  # same order
         print('Reversing anchor order')
         m.anchors[:] = m.anchors.flip(0)
-        m.anchor_grid[:] = m.anchor_grid.flip(0)
 
 
 def check_anchors(dataset, model, thr=4.0, imgsz=640):
@@ -41,12 +40,12 @@ def check_anchors(dataset, model, thr=4.0, imgsz=640):
         bpr = (best > 1. / thr).float().mean()  # best possible recall
         return bpr, aat
 
-    anchors = m.anchor_grid.clone().cpu().view(-1, 2)  # current anchors
-    bpr, aat = metric(anchors)
+    anchors = m.anchors.clone() * m.stride.to(m.anchors.device).view(-1, 1, 1)  # current anchors
+    bpr, aat = metric(anchors.cpu().view(-1, 2))
     print(f'anchors/target = {aat:.2f}, Best Possible Recall (BPR) = {bpr:.4f}', end='')
     if bpr < 0.98:  # threshold to recompute
         print('. Attempting to improve anchors, please wait...')
-        na = m.anchor_grid.numel() // 2  # number of anchors
+        na = m.anchors.numel() // 2  # number of anchors
         try:
             anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False)
         except Exception as e:
@@ -54,7 +53,6 @@ def check_anchors(dataset, model, thr=4.0, imgsz=640):
         new_bpr = metric(anchors)[0]
         if new_bpr > bpr:  # replace anchors
             anchors = torch.tensor(anchors, device=m.anchors.device).type_as(m.anchors)
-            m.anchor_grid[:] = anchors.clone().view_as(m.anchor_grid)  # for inference
             m.anchors[:] = anchors.clone().view_as(m.anchors) / m.stride.to(m.anchors.device).view(-1, 1, 1)  # loss
             check_anchor_order(m)
             print(f'{prefix}New anchors saved to model. Update model *.yaml to use these anchors in the future.')