Use C3 following the model specification of YOLOv5 (#343)

* Use C3 following the model specification of YOLOv5

* Add unittest for yolov5_mobilenet_v3_small_fpn

* Add type annotations

* Minor fix for weights_name

test/test_models.py

@@ -12,6 +12,7 @@
 from yolort.models.backbone_utils import darknet_pan_backbone
 from yolort.models.box_head import YOLOHead, PostProcess, SetCriterion
 from yolort.models.transformer import darknet_tan_backbone
+from yolort.models.yolo_lite import yolov5_mobilenet_v3_small_fpn
 from yolort.v5 import get_yolov5_size, attempt_download
 
 
@@ -420,3 +421,18 @@ def test_load_from_yolov5_torchscript(arch, size_divisible, version, upstream_ve
     torch.testing.assert_close(out[0]["scores"], out_script[1][0]["scores"], rtol=0, atol=0)
     torch.testing.assert_close(out[0]["labels"], out_script[1][0]["labels"], rtol=0, atol=0)
     torch.testing.assert_close(out[0]["boxes"], out_script[1][0]["boxes"], rtol=0, atol=0)
+
+
+def test_yolov5_mobilenet_v3_small_fpn():
+
+    model = yolov5_mobilenet_v3_small_fpn()
+    model = model.eval()
+
+    images = torch.rand(4, 3, 320, 320)
+    out = model(images)
+    assert isinstance(out, list)
+    assert len(out) == 4
+    assert isinstance(out[0], dict)
+    assert isinstance(out[0]["boxes"], Tensor)
+    assert isinstance(out[0]["labels"], Tensor)
+    assert isinstance(out[0]["scores"], Tensor)

yolort/models/box_head.py

@@ -20,7 +20,7 @@ def __init__(
     ):
         super().__init__()
         if not isinstance(in_channels, list):
-            in_channels = [in_channels] * num_anchors
+            in_channels = [in_channels] * len(strides)
         self.num_anchors = num_anchors  # anchors
         self.num_classes = num_classes
         self.num_outputs = num_classes + 5  # number of outputs per anchor

yolort/models/yolo_lite.py

@@ -1,10 +1,15 @@
+from typing import Dict, List, Callable, Optional
+
 from torch import nn
 from torchvision.models import mobilenet
 from torchvision.models._utils import IntermediateLayerGetter
 from torchvision.models.detection.backbone_utils import _validate_trainable_layers
 from torchvision.ops import misc as misc_nn_ops
-from torchvision.ops.feature_pyramid_network import FeaturePyramidNetwork, LastLevelMaxPool
+from torchvision.ops.feature_pyramid_network import ExtraFPNBlock, FeaturePyramidNetwork, LastLevelMaxPool
+from yolort.utils import load_state_dict_from_url
 
+from .anchor_utils import AnchorGenerator
+from .box_head import YOLOHead
 from .yolo import YOLO
 
 __all__ = ["yolov5_mobilenet_v3_small_fpn"]
@@ -31,7 +36,14 @@ class BackboneWithFPN(nn.Module):
         out_channels (int): the number of channels in the FPN
     """
 
-    def __init__(self, backbone, return_layers, in_channels_list, out_channels, extra_blocks=None):
+    def __init__(
+        self,
+        backbone: nn.Module,
+        return_layers: Dict[str, str],
+        in_channels_list: List[int],
+        out_channels: int,
+        extra_blocks: Optional[ExtraFPNBlock] = None,
+    ) -> None:
         super().__init__()
 
         if extra_blocks is None:
@@ -53,12 +65,12 @@ def forward(self, x):
 
 
 def mobilenet_backbone(
-    backbone_name,
-    pretrained,
-    norm_layer=misc_nn_ops.FrozenBatchNorm2d,
-    trainable_layers=2,
-    returned_layers=None,
-):
+    backbone_name: str,
+    pretrained: bool,
+    norm_layer: Callable[..., nn.Module] = misc_nn_ops.FrozenBatchNorm2d,
+    trainable_layers: int = 2,
+    returned_layers: Optional[List[int]] = None,
+) -> nn.Module:
     backbone = mobilenet.__dict__[backbone_name](pretrained=pretrained, norm_layer=norm_layer).features
 
     # Gather the indices of blocks which are strided. These are the locations of C1, ..., Cn-1 blocks.
@@ -79,7 +91,7 @@ def mobilenet_backbone(
     out_channels = 256
 
     if returned_layers is None:
-        returned_layers = [num_stages - 2, num_stages - 1]
+        returned_layers = [num_stages - 3, num_stages - 2, num_stages - 1]
     assert min(returned_layers) >= 0 and max(returned_layers) < num_stages
     return_layers = {f"{stage_indices[k]}": str(v) for v, k in enumerate(returned_layers)}
 
@@ -94,12 +106,12 @@ def mobilenet_backbone(
 
 
 def _yolov5_mobilenet_v3_small_fpn(
-    weights_name,
-    pretrained=False,
-    progress=True,
-    num_classes=80,
-    pretrained_backbone=True,
-    trainable_backbone_layers=None,
+    weights_name: str,
+    pretrained: bool = False,
+    progress: bool = True,
+    num_classes: int = 80,
+    pretrained_backbone: bool = True,
+    trainable_backbone_layers: Optional[int] = None,
     **kwargs,
 ):
     trainable_backbone_layers = _validate_trainable_layers(
@@ -113,18 +125,40 @@ def _yolov5_mobilenet_v3_small_fpn(
         pretrained_backbone,
         trainable_layers=trainable_backbone_layers,
     )
+    strides = [8, 16, 32, 64]
+    anchor_grids = [
+        [19, 27, 44, 40, 38, 94],
+        [96, 68, 86, 152, 180, 137],
+        [140, 301, 303, 264, 238, 542],
+        [436, 615, 739, 380, 925, 792],
+    ]
+    anchor_generator = AnchorGenerator(strides, anchor_grids)
+    head = YOLOHead(
+        backbone.out_channels,
+        anchor_generator.num_anchors,
+        anchor_generator.strides,
+        num_classes,
+    )
+    model = YOLO(backbone, num_classes, anchor_generator=anchor_generator, head=head, **kwargs)
+    if pretrained:
+        if model_urls.get(weights_name, None) is None:
+            raise ValueError(f"No checkpoint is available for model {weights_name}")
+        state_dict = load_state_dict_from_url(model_urls[weights_name], progress=progress)
+        model.load_state_dict(state_dict)
+    return model
 
-    model = YOLO(backbone, num_classes, **kwargs)
 
-    return model
+model_urls = {
+    "yolov5_mobilenet_v3_small_fpn_coco": None,
+}
 
 
 def yolov5_mobilenet_v3_small_fpn(
-    pretrained=False,
-    progress=True,
-    num_classes=80,
-    pretrained_backbone=True,
-    trainable_backbone_layers=None,
+    pretrained: bool = False,
+    progress: bool = True,
+    num_classes: int = 80,
+    pretrained_backbone: bool = True,
+    trainable_backbone_layers: Optional[int] = None,
     **kwargs,
 ):
     """