From c205f7ec617b90c92872628db62ab8661bcfd079 Mon Sep 17 00:00:00 2001
From: Florian Vahl <7vahl@informatik.uni-hamburg.de>
Date: Sat, 23 Nov 2024 16:00:30 +0100
Subject: [PATCH 1/3] Add deconvolution

---
 yoeo/models.py | 37 ++++++++++++++++++++++++++++++++++---
 1 file changed, 34 insertions(+), 3 deletions(-)

diff --git a/yoeo/models.py b/yoeo/models.py
index 1b69cd8..cd964c0 100644
--- a/yoeo/models.py
+++ b/yoeo/models.py
@@ -60,8 +60,40 @@ def create_modules(module_defs):
                                    nn.BatchNorm2d(filters, momentum=0.1, eps=1e-5))
             if module_def["activation"] == "leaky":
                 modules.add_module(f"leaky_{module_i}", nn.LeakyReLU(0.1))
-            if module_def["activation"] == "mish":
+            elif module_def["activation"] == "mish":
                 modules.add_module(f"mish_{module_i}", Mish())
+            elif module_def["activation"] == "linear":
+                pass
+            else:
+                raise ValueError(f"Unknown activation: {module_def['activation']}")
+
+        elif module_def["type"] == "deconvolutional":
+            bn = int(module_def["batch_normalize"])
+            filters = int(module_def["filters"])
+            kernel_size = int(module_def["size"])
+            pad = int(module_def["pad"])
+            modules.add_module(
+                f"deconv_{module_i}",
+                nn.ConvTranspose2d(
+                    in_channels=output_filters[-1],
+                    out_channels=filters,
+                    kernel_size=kernel_size,
+                    stride=int(module_def["stride"]),
+                    padding=pad,
+                    bias=not bn,
+                ),
+            )
+            if bn:
+                modules.add_module(f"batch_norm_{module_i}",
+                                   nn.BatchNorm2d(filters, momentum=0.1, eps=1e-5))
+            if module_def["activation"] == "leaky":
+                modules.add_module(f"leaky_{module_i}", nn.LeakyReLU(0.1))
+            elif module_def["activation"] == "mish":
+                modules.add_module(f"mish_{module_i}", Mish())
+            elif module_def["activation"] == "linear":
+                pass
+            else:
+                raise ValueError(f"Unknown activation: {module_def['activation']}")
 
         elif module_def["type"] == "maxpool":
             kernel_size = int(module_def["size"])
@@ -197,10 +229,9 @@ def __init__(self, config_path):
 
     def forward(self, x, bb_targets=None, mask_targets=None):
         img_size = x.size(2)
-        loss = 0
         layer_outputs, yolo_outputs, segmentation_outputs = [], [], []
         for i, (module_def, module) in enumerate(zip(self.module_defs, self.module_list)):
-            if module_def["type"] in ["convolutional", "upsample", "maxpool"]:
+            if module_def["type"] in ["convolutional", "deconvolutional", "upsample", "maxpool"]:
                 x = module(x)
             elif module_def["type"] == "route":
                 combined_outputs = torch.cat([layer_outputs[int(layer_i)] for layer_i in module_def["layers"].split(",")], 1)

From dbe237272112e82d2c03047c46484f30243014a4 Mon Sep 17 00:00:00 2001
From: Florian Vahl <7vahl@informatik.uni-hamburg.de>
Date: Sat, 23 Nov 2024 16:00:53 +0100
Subject: [PATCH 2/3] Add model variations

---
 config/yoeo_light_decoder_deconv.cfg  | 312 ++++++++++++++++++++++
 config/yoeo_medium_decoder_deconv.cfg | 356 ++++++++++++++++++++++++++
 2 files changed, 668 insertions(+)
 create mode 100644 config/yoeo_light_decoder_deconv.cfg
 create mode 100644 config/yoeo_medium_decoder_deconv.cfg

diff --git a/config/yoeo_light_decoder_deconv.cfg b/config/yoeo_light_decoder_deconv.cfg
new file mode 100644
index 0000000..f0b551e
--- /dev/null
+++ b/config/yoeo_light_decoder_deconv.cfg
@@ -0,0 +1,312 @@
+[net]
+# Testing
+#batch=1
+#subdivisions=1
+# Training
+batch=64
+subdivisions=4
+width=416
+height=416
+channels=3
+momentum=0.9
+decay=0.0005
+angle=0
+saturation = 1.5
+exposure = 1.5
+hue=.1
+
+learning_rate=0.0001
+burn_in=100
+max_batches = 4000
+policy=steps
+steps=50000,60000
+scales=.1,.1
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+###########
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=8
+activation=linear
+
+[yolo]
+mask = 0
+anchors = 100, 100
+classes=3
+num=1
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=1
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[upsample]
+stride=2
+
+[route]
+layers = -1, 24
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=8
+activation=linear
+
+[yolo]
+mask = 0
+anchors = 100, 100
+classes=3
+num=1
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=1
+
+[route]
+layers = 18
+
+[deconvolutional]
+batch_normalize=1
+filters=128
+size=2
+stride=2
+pad=0
+activation=leaky
+
+[route]
+layers = -1, 10
+
+[deconvolutional]
+batch_normalize=1
+filters=64
+size=2
+stride=2
+pad=0
+activation=leaky
+
+[route]
+layers = -1, 2
+
+[deconvolutional]
+batch_normalize=1
+filters=32
+size=2
+stride=2
+pad=0
+activation=leaky
+
+[route]
+layers = -1, 0
+
+[deconvolutional]
+batch_normalize=1
+filters=16
+size=2
+stride=2
+pad=0
+activation=linear
+
+[seg]
+classes=3
diff --git a/config/yoeo_medium_decoder_deconv.cfg b/config/yoeo_medium_decoder_deconv.cfg
new file mode 100644
index 0000000..45df2c2
--- /dev/null
+++ b/config/yoeo_medium_decoder_deconv.cfg
@@ -0,0 +1,356 @@
+[net]
+# Testing
+#batch=1
+#subdivisions=1
+# Training
+batch=64
+subdivisions=4
+width=416
+height=416
+channels=3
+momentum=0.9
+decay=0.0005
+angle=0
+saturation = 1.5
+exposure = 1.5
+hue=.1
+
+learning_rate=0.0001
+burn_in=100
+max_batches = 4000
+policy=steps
+steps=50000,60000
+scales=.1,.1
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+###########
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=8
+activation=linear
+
+[yolo]
+mask = 0
+anchors = 100, 100
+classes=3
+num=1
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=1
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[upsample]
+stride=2
+
+[route]
+layers = -1, 24
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=8
+activation=linear
+
+[yolo]
+mask = 0
+anchors = 100, 100
+classes=3
+num=1
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=1
+
+[route]
+layers = 18
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[shortcut]
+from=-2
+
+[deconvolutional]
+batch_normalize=1
+filters=128
+size=2
+stride=2
+pad=0
+activation=leaky
+
+[route]
+layers = -1, 10
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[shortcut]
+from=-2
+
+[deconvolutional]
+batch_normalize=1
+filters=64
+size=2
+stride=2
+pad=0
+activation=leaky
+
+[route]
+layers = -1, 2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[shortcut]
+from=-2
+
+[deconvolutional]
+batch_normalize=1
+filters=32
+size=2
+stride=2
+pad=0
+activation=leaky
+
+[route]
+layers = -1, 0
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[shortcut]
+from=-2
+
+[deconvolutional]
+batch_normalize=1
+filters=3
+size=2
+stride=2
+pad=0
+activation=linear
+
+[seg]
+classes=3

From 84e12674cc1726f9bbf488c091d0e9c709a34a32 Mon Sep 17 00:00:00 2001
From: Florian Vahl <7vahl@informatik.uni-hamburg.de>
Date: Wed, 18 Dec 2024 20:51:53 +0100
Subject: [PATCH 3/3] Add version with large max pool

---
 config/yoeo_medium_decoder_deconv_context.cfg | 381 ++++++++++++++++++
 1 file changed, 381 insertions(+)
 create mode 100644 config/yoeo_medium_decoder_deconv_context.cfg

diff --git a/config/yoeo_medium_decoder_deconv_context.cfg b/config/yoeo_medium_decoder_deconv_context.cfg
new file mode 100644
index 0000000..cd1803d
--- /dev/null
+++ b/config/yoeo_medium_decoder_deconv_context.cfg
@@ -0,0 +1,381 @@
+[net]
+# Testing
+#batch=1
+#subdivisions=1
+# Training
+batch=64
+subdivisions=4
+width=416
+height=416
+channels=3
+momentum=0.9
+decay=0.0005
+angle=0
+saturation = 1.5
+exposure = 1.5
+hue=.1
+
+learning_rate=0.0001
+burn_in=100
+max_batches = 4000
+policy=steps
+steps=50000,60000
+scales=.1,.1
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+###########
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=8
+activation=linear
+
+[yolo]
+mask = 0
+anchors = 100, 100
+classes=3
+num=1
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=1
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[upsample]
+stride=2
+
+[route]
+layers = -1, 24
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=8
+activation=linear
+
+[yolo]
+mask = 0
+anchors = 100, 100
+classes=3
+num=1
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=1
+
+[route]
+layers = 18
+groups=2
+group_id=1
+
+[maxpool]
+size=13
+stride=13
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[upsample]
+stride=13
+
+[route]
+layers = 18
+groups=2
+group_id=0
+
+[route]
+layers = -1, -2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[shortcut]
+from=-2
+
+[deconvolutional]
+batch_normalize=1
+filters=128
+size=2
+stride=2
+pad=0
+activation=leaky
+
+[route]
+layers = -1, 10
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[shortcut]
+from=-2
+
+[deconvolutional]
+batch_normalize=1
+filters=64
+size=2
+stride=2
+pad=0
+activation=leaky
+
+[route]
+layers = -1, 2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[shortcut]
+from=-2
+
+[deconvolutional]
+batch_normalize=1
+filters=32
+size=2
+stride=2
+pad=0
+activation=leaky
+
+[route]
+layers = -1, 0
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[shortcut]
+from=-2
+
+[deconvolutional]
+batch_normalize=1
+filters=3
+size=2
+stride=2
+pad=0
+activation=linear
+
+[seg]
+classes=3