ultralytics 8.2.35 add YOLOv9t/s/m models (ultralytics#13504)

Co-authored-by: UltralyticsAssistant <[email protected]> Co-authored-by: Glenn Jocher <[email protected]>
Burhan-Q · Jun 17, 2024 · 8ea945c · 8ea945c
1 parent 605e7f4
commit 8ea945c
Show file tree

Hide file tree

Showing 10 changed files with 168 additions and 8 deletions.
diff --git a/docs/en/models/yolov9.md b/docs/en/models/yolov9.md
@@ -79,9 +79,9 @@ The performance of YOLOv9 on the [COCO dataset](../datasets/detect/coco.md) exem
 
         | Model                                                                                 | size<br><sup>(pixels) | mAP<sup>val<br>50-95 | mAP<sup>val<br>50 | params<br><sup>(M) | FLOPs<br><sup>(B) |
         |---------------------------------------------------------------------------------------|-----------------------|----------------------|-------------------|--------------------|-------------------|
-        | YOLOv9t                                                                               | 640                   | 38.3                 | 53.1              | 2.0                | 7.7               |
-        | YOLOv9s                                                                               | 640                   | 46.8                 | 63.4              | 7.2                | 26.7              |
-        | YOLOv9m                                                                               | 640                   | 51.4                 | 68.1              | 20.1               | 76.8              |
+        | [YOLOv9t](https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov9t.pt)  | 640                   | 38.3                 | 53.1              | 2.0                | 7.7               |
+        | [YOLOv9s](https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov9s.pt)  | 640                   | 46.8                 | 63.4              | 7.2                | 26.7              |
+        | [YOLOv9m](https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov9m.pt)  | 640                   | 51.4                 | 68.1              | 20.1               | 76.8              |
         | [YOLOv9c](https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov9c.pt)  | 640                   | 53.0                 | 70.2              | 25.5               | 102.8             |
         | [YOLOv9e](https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov9e.pt)  | 640                   | 55.6                 | 72.8              | 58.1               | 192.5             |
 
@@ -152,10 +152,10 @@ This example provides simple YOLOv9 training and inference examples. For full do
 
 The YOLOv9 series offers a range of models, each optimized for high-performance [Object Detection](../tasks/detect.md). These models cater to varying computational needs and accuracy requirements, making them versatile for a wide array of applications.
 
-| Model      | Filenames                         | Tasks                                        | Inference | Validation | Training | Export |
-| ---------- | --------------------------------- | -------------------------------------------- | --------- | ---------- | -------- | ------ |
-| YOLOv9     | `yolov9c.pt` `yolov9e.pt`         | [Object Detection](../tasks/detect.md)       | ✅        | ✅         | ✅       | ✅     |
-| YOLOv9-seg | `yolov9c-seg.pt` `yolov9e-seg.pt` | [Instance Segmentation](../tasks/segment.md) | ✅        | ✅         | ✅       | ✅     |
+| Model      | Filenames                                               | Tasks                                        | Inference | Validation | Training | Export |
+|------------|---------------------------------------------------------|----------------------------------------------|-----------|------------|----------|--------|
+| YOLOv9     | `yolov9t` `yolov9s` `yolov9m` `yolov9c.pt` `yolov9e.pt` | [Object Detection](../tasks/detect.md)       | ✅        | ✅         | ✅       | ✅     |
+| YOLOv9-seg | `yolov9c-seg.pt` `yolov9e-seg.pt`                       | [Instance Segmentation](../tasks/segment.md) | ✅        | ✅         | ✅       | ✅     |
 
 This table provides a detailed overview of the YOLOv9 model variants, highlighting their capabilities in object detection tasks and their compatibility with various operational modes such as [Inference](../modes/predict.md), [Validation](../modes/val.md), [Training](../modes/train.md), and [Export](../modes/export.md). This comprehensive support ensures that users can fully leverage the capabilities of YOLOv9 models in a broad range of object detection scenarios.
 

diff --git a/docs/en/reference/nn/modules/block.md b/docs/en/reference/nn/modules/block.md
@@ -119,6 +119,14 @@ keywords: Ultralytics, YOLO, neural networks, block modules, DFL, Proto, HGStem,
 
 <br><br>
 
+## ::: ultralytics.nn.modules.block.ELAN1
+
+<br><br>
+
+## ::: ultralytics.nn.modules.block.AConv
+
+<br><br>
+
 ## ::: ultralytics.nn.modules.block.ADown
 
 <br><br>

diff --git a/tests/test_cuda.py b/tests/test_cuda.py
@@ -19,6 +19,7 @@ def test_checks():
 
 
 @pytest.mark.slow
+@pytest.mark.skipif(True, reason="CUDA export tests disabled pending additional Ultralytics GPU server availability")
 @pytest.mark.skipif(not CUDA_IS_AVAILABLE, reason="CUDA is not available")
 @pytest.mark.parametrize(
     "task, dynamic, int8, half, batch",

diff --git a/ultralytics/__init__.py b/ultralytics/__init__.py
@@ -1,6 +1,6 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
 
-__version__ = "8.2.34"
+__version__ = "8.2.35"
 
 import os
 

diff --git a/ultralytics/cfg/models/v9/yolov9m.yaml b/ultralytics/cfg/models/v9/yolov9m.yaml
@@ -0,0 +1,38 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+# YOLOv9t
+# 603 layers, 20216160 parameters, 77.9 GFLOPs
+
+# parameters
+nc: 80 # number of classes
+
+# gelan backbone
+backbone:
+  - [-1, 1, Conv, [32, 3, 2]] # 0-P1/2
+  - [-1, 1, Conv, [64, 3, 2]] # 1-P2/4
+  - [-1, 1, RepNCSPELAN4, [128, 128, 64, 1]] # 2
+  - [-1, 1, AConv, [240]] # 3-P3/8
+  - [-1, 1, RepNCSPELAN4, [240, 240, 120, 1]] # 4
+  - [-1, 1, AConv, [360]] # 5-P4/16
+  - [-1, 1, RepNCSPELAN4, [360, 360, 180, 1]] # 6
+  - [-1, 1, AConv, [480]] # 7-P5/32
+  - [-1, 1, RepNCSPELAN4, [480, 480, 240, 1]] # 8
+  - [-1, 1, SPPELAN, [480, 240]] # 9
+
+head:
+  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
+  - [[-1, 6], 1, Concat, [1]] # cat backbone P4
+  - [-1, 1, RepNCSPELAN4, [360, 360, 180, 1]] # 12
+
+  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
+  - [[-1, 4], 1, Concat, [1]] # cat backbone P3
+  - [-1, 1, RepNCSPELAN4, [240, 240, 120, 1]] # 15
+
+  - [-1, 1, AConv, [180]]
+  - [[-1, 12], 1, Concat, [1]] # cat head P4
+  - [-1, 1, RepNCSPELAN4, [360, 360, 180, 1]] # 18 (P4/16-medium)
+
+  - [-1, 1, AConv, [240]]
+  - [[-1, 9], 1, Concat, [1]] # cat head P5
+  - [-1, 1, RepNCSPELAN4, [480, 480, 240, 1]] # 21 (P5/32-large)
+
+  - [[15, 18, 21], 1, Detect, [nc]] # Detect(P3, P4, P5)
diff --git a/ultralytics/cfg/models/v9/yolov9s.yaml b/ultralytics/cfg/models/v9/yolov9s.yaml
@@ -0,0 +1,38 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+# YOLOv9s
+# 917 layers, 7318368 parameters, 27.6 GFLOPs
+
+# parameters
+nc: 80 # number of classes
+
+# gelan backbone
+backbone:
+  - [-1, 1, Conv, [32, 3, 2]] # 0-P1/2
+  - [-1, 1, Conv, [64, 3, 2]] # 1-P2/4
+  - [-1, 1, ELAN1, [64, 64, 32]] # 2
+  - [-1, 1, AConv, [128]] # 3-P3/8
+  - [-1, 1, RepNCSPELAN4, [128, 128, 64, 3]] # 4
+  - [-1, 1, AConv, [192]] # 5-P4/16
+  - [-1, 1, RepNCSPELAN4, [192, 192, 96, 3]] # 6
+  - [-1, 1, AConv, [256]] # 7-P5/32
+  - [-1, 1, RepNCSPELAN4, [256, 256, 128, 3]] # 8
+  - [-1, 1, SPPELAN, [256, 128]] # 9
+
+head:
+  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
+  - [[-1, 6], 1, Concat, [1]] # cat backbone P4
+  - [-1, 1, RepNCSPELAN4, [192, 192, 96, 3]] # 12
+
+  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
+  - [[-1, 4], 1, Concat, [1]] # cat backbone P3
+  - [-1, 1, RepNCSPELAN4, [128, 128, 64, 3]] # 15
+
+  - [-1, 1, AConv, [96]]
+  - [[-1, 12], 1, Concat, [1]] # cat head P4
+  - [-1, 1, RepNCSPELAN4, [192, 192, 96, 3]] # 18 (P4/16-medium)
+
+  - [-1, 1, AConv, [128]]
+  - [[-1, 9], 1, Concat, [1]] # cat head P5
+  - [-1, 1, RepNCSPELAN4, [256, 256, 128, 3]] # 21 (P5/32-large)
+
+  - [[15, 18, 21], 1, Detect, [nc]] # Detect(P3, P4 P5)
diff --git a/ultralytics/cfg/models/v9/yolov9t.yaml b/ultralytics/cfg/models/v9/yolov9t.yaml
@@ -0,0 +1,38 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+# YOLOv9t
+# 917 layers, 2128720 parameters, 8.5 GFLOPs
+
+# parameters
+nc: 80 # number of classes
+
+# gelan backbone
+backbone:
+  - [-1, 1, Conv, [16, 3, 2]] # 0-P1/2
+  - [-1, 1, Conv, [32, 3, 2]] # 1-P2/4
+  - [-1, 1, ELAN1, [32, 32, 16]] # 2
+  - [-1, 1, AConv, [64]] # 3-P3/8
+  - [-1, 1, RepNCSPELAN4, [64, 64, 32, 3]] # 4
+  - [-1, 1, AConv, [96]] # 5-P4/16
+  - [-1, 1, RepNCSPELAN4, [96, 96, 48, 3]] # 6
+  - [-1, 1, AConv, [128]] # 7-P5/32
+  - [-1, 1, RepNCSPELAN4, [128, 128, 64, 3]] # 8
+  - [-1, 1, SPPELAN, [128, 64]] # 9
+
+head:
+  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
+  - [[-1, 6], 1, Concat, [1]] # cat backbone P4
+  - [-1, 1, RepNCSPELAN4, [96, 96, 48, 3]] # 12
+
+  - [-1, 1, nn.Upsample, [None, 2, "nearest"]]
+  - [[-1, 4], 1, Concat, [1]] # cat backbone P3
+  - [-1, 1, RepNCSPELAN4, [64, 64, 32, 3]] # 15
+
+  - [-1, 1, AConv, [48]]
+  - [[-1, 12], 1, Concat, [1]] # cat head P4
+  - [-1, 1, RepNCSPELAN4, [96, 96, 48, 3]] # 18 (P4/16-medium)
+
+  - [-1, 1, AConv, [64]]
+  - [[-1, 9], 1, Concat, [1]] # cat head P5
+  - [-1, 1, RepNCSPELAN4, [128, 128, 64, 3]] # 21 (P5/32-large)
+
+  - [[15, 18, 21], 1, Detect, [nc]] # Detect(P3, P4, P5)
diff --git a/ultralytics/nn/modules/__init__.py b/ultralytics/nn/modules/__init__.py
@@ -23,9 +23,11 @@
     C3,
     C3TR,
     DFL,
+    ELAN1,
     SPP,
     SPPELAN,
     SPPF,
+    AConv,
     ADown,
     BNContrastiveHead,
     Bottleneck,
@@ -133,4 +135,6 @@
     "SPPELAN",
     "CBFuse",
     "CBLinear",
+    "AConv",
+    "ELAN1",
 )
diff --git a/ultralytics/nn/modules/block.py b/ultralytics/nn/modules/block.py
@@ -32,7 +32,9 @@
     "RepC3",
     "ResNetLayer",
     "RepNCSPELAN4",
+    "ELAN1",
     "ADown",
+    "AConv",
     "SPPELAN",
     "CBFuse",
     "CBLinear",
@@ -603,6 +605,33 @@ def forward_split(self, x):
         return self.cv4(torch.cat(y, 1))
 
 
+class ELAN1(RepNCSPELAN4):
+    """ELAN1 module with 4 convolutions."""
+
+    def __init__(self, c1, c2, c3, c4):
+        """Initializes ELAN1 layer with specified channel sizes."""
+        super().__init__(c1, c2, c3, c4)
+        self.c = c3 // 2
+        self.cv1 = Conv(c1, c3, 1, 1)
+        self.cv2 = Conv(c3 // 2, c4, 3, 1)
+        self.cv3 = Conv(c4, c4, 3, 1)
+        self.cv4 = Conv(c3 + (2 * c4), c2, 1, 1)
+
+
+class AConv(nn.Module):
+    """AConv."""
+
+    def __init__(self, c1, c2):
+        """Initializes AConv module with convolution layers."""
+        super().__init__()
+        self.cv1 = Conv(c1, c2, 3, 2, 1)
+
+    def forward(self, x):
+        """Forward pass through AConv layer."""
+        x = torch.nn.functional.avg_pool2d(x, 2, 1, 0, False, True)
+        return self.cv1(x)
+
+
 class ADown(nn.Module):
     """ADown."""
 

diff --git a/ultralytics/nn/tasks.py b/ultralytics/nn/tasks.py
@@ -13,10 +13,12 @@
     C2,
     C3,
     C3TR,
+    ELAN1,
     OBB,
     SPP,
     SPPELAN,
     SPPF,
+    AConv,
     ADown,
     Bottleneck,
     BottleneckCSP,
@@ -878,7 +880,9 @@ def parse_model(d, ch, verbose=True):  # model_dict, input_channels(3)
             C2,
             C2f,
             RepNCSPELAN4,
+            ELAN1,
             ADown,
+            AConv,
             SPPELAN,
             C2fAttn,
             C3,