Per-stage eca MV3 config, going to shelve these experiments for a while, not proving very productive

timm/models/efficientnet_builder.py

@@ -85,7 +85,8 @@ def _decode_block_str(block_str):
         attn_layer = 'sev2'
         attn_kwargs = dict(se_ratio=float(options['se']))
     elif 'eca' in options:
-        attn_layer = 'eca'
+        attn_layer = 'ceca'
+        attn_kwargs = dict(kernel_size=int(options['eca']))
 
     num_repeat = int(options['r'])
     # each type of block has different valid arguments, fill accordingly
@@ -378,7 +379,7 @@ class EfficientNetBuilder:
         return stages
 
 
-def _init_weight_goog(m, n='', fix_group_fanout=False):
+def _init_weight_goog(m, n='', fix_group_fanout=True):
     """ Weight initialization as per Tensorflow official implementations.
 
     Args:

timm/models/mobilenetv3.py

@@ -389,23 +389,38 @@ def _gen_mobilenet_v3_eca(variant, channel_multiplier=1.0, pretrained=False, **k
     else:
         num_features = 1280
         act_layer = HardSwish
+        # arch_def = [
+        #     # stage 0, 112x112 in
+        #     ['ds_r1_k3_s1_e1_c16_nre'],  # relu
+        #     # stage 1, 112x112 in
+        #     ['ir_r1_k3_s2_e4_c24_nre', 'ir_r1_k3_s1_e3_c24_nre'],  # relu
+        #     # stage 2, 56x56 in
+        #     ['ir_r3_k5_s2_e3_c40_nre'],  # relu
+        #     # stage 3, 28x28 in
+        #     ['ir_r1_k3_s2_e6_c80', 'ir_r1_k3_s1_e2.5_c80', 'ir_r2_k3_s1_e2.3_c80'],  # hard-swish
+        #     # stage 4, 14x14in
+        #     ['ir_r2_k3_s1_e6_c112'],  # hard-swish
+        #     # stage 5, 14x14in
+        #     ['ir_r3_k5_s2_e6_c160'],  # hard-swish
+        #     # stage 6, 7x7 in
+        #     ['cn_r1_k1_s1_c960'],  # hard-swish
+        # ]
         arch_def = [
             # stage 0, 112x112 in
             ['ds_r1_k3_s1_e1_c16_nre'],  # relu
             # stage 1, 112x112 in
             ['ir_r1_k3_s2_e4_c24_nre', 'ir_r1_k3_s1_e3_c24_nre'],  # relu
             # stage 2, 56x56 in
-            ['ir_r3_k5_s2_e3_c40_nre'],  # relu
+            ['ir_r3_k5_s2_e3_c40_eca3_nre'],  # relu
             # stage 3, 28x28 in
-            ['ir_r1_k3_s2_e6_c80', 'ir_r1_k3_s1_e2.5_c80', 'ir_r2_k3_s1_e2.3_c80'],  # hard-swish
+            ['ir_r1_k3_s2_e6_c80_eca3', 'ir_r1_k3_s1_e2.5_c80_eca3', 'ir_r2_k3_s1_e2.3_c80_eca3'],  # hard-swish
             # stage 4, 14x14in
-            ['ir_r2_k3_s1_e6_c112'],  # hard-swish
+            ['ir_r2_k3_s1_e6_c112_eca5'],  # hard-swish
             # stage 5, 14x14in
-            ['ir_r3_k5_s2_e6_c160'],  # hard-swish
+            ['ir_r3_k5_s2_e6_c160_eca5'],  # hard-swish
             # stage 6, 7x7 in
             ['cn_r1_k1_s1_c960'],  # hard-swish
         ]
-
     model_kwargs = dict(
         block_args=decode_arch_def(arch_def),
         num_features=num_features,
@@ -413,7 +428,7 @@ def _gen_mobilenet_v3_eca(variant, channel_multiplier=1.0, pretrained=False, **k
         channel_multiplier=channel_multiplier,
         norm_kwargs=resolve_bn_args(kwargs),
         act_layer=act_layer,
-        attn_layer='ceca',
+        #attn_layer='ceca',
         attn_kwargs=dict(gate_fn=hard_sigmoid),
         **kwargs,
     )