Make gcvit window size ratio based to improve resolution changing support #1449. Change default init to original.

Author: rwightman

timm/models/gcvit.py

@@ -30,8 +30,8 @@
 from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
 from .fx_features import register_notrace_function
 from .helpers import build_model_with_cfg, named_apply
-from .layers import trunc_normal_tf_, DropPath, to_2tuple, Mlp, get_attn, get_act_layer, get_norm_layer, \
-    ClassifierHead, LayerNorm2d, _assert
+from .layers import DropPath, to_2tuple, to_ntuple, Mlp, ClassifierHead, LayerNorm2d,\
+    get_attn, get_act_layer, get_norm_layer, _assert
 from .registry import register_model
 from .vision_transformer_relpos import RelPosMlp, RelPosBias  # FIXME move to common location
 
@@ -321,7 +321,7 @@ def __init__(
             depth: int,
             num_heads: int,
             feat_size: Tuple[int, int],
-            window_size: int,
+            window_size: Tuple[int, int],
             downsample: bool = True,
             global_norm: bool = False,
             stage_norm: bool = False,
@@ -347,8 +347,9 @@ def __init__(
         else:
             self.downsample = nn.Identity()
         self.feat_size = feat_size
+        window_size = to_2tuple(window_size)
 
-        feat_levels = int(math.log2(min(feat_size) / window_size))
+        feat_levels = int(math.log2(min(feat_size) / min(window_size)))
         self.global_block = FeatureBlock(dim, feat_levels)
         self.global_norm = norm_layer_cl(dim) if global_norm else nn.Identity()
 
@@ -400,7 +401,8 @@ def __init__(
             num_classes: int = 1000,
             global_pool: str = 'avg',
             img_size: Tuple[int, int] = 224,
-            window_size: Tuple[int, ...] = (7, 7, 14, 7),
+            window_ratio: Tuple[int, ...] = (32, 32, 16, 32),
+            window_size: Tuple[int, ...] = None,
             embed_dim: int = 64,
             depths: Tuple[int, ...] = (3, 4, 19, 5),
             num_heads: Tuple[int, ...] = (2, 4, 8, 16),
@@ -411,7 +413,7 @@ def __init__(
             proj_drop_rate: float = 0.,
             attn_drop_rate: float = 0.,
             drop_path_rate: float = 0.,
-            weight_init='vit',
+            weight_init='',
             act_layer: str = 'gelu',
             norm_layer: str = 'layernorm2d',
             norm_layer_cl: str = 'layernorm',
@@ -429,6 +431,11 @@ def __init__(
         self.drop_rate = drop_rate
         num_stages = len(depths)
         self.num_features = int(embed_dim * 2 ** (num_stages - 1))
+        if window_size is not None:
+            window_size = to_ntuple(num_stages)(window_size)
+        else:
+            assert window_ratio is not None
+            window_size = tuple([(img_size[0] // r, img_size[1] // r) for r in to_ntuple(num_stages)(window_ratio)])
 
         self.stem = Stem(
             in_chs=in_chans,
@@ -480,7 +487,7 @@ def _init_weights(self, module, name, scheme='vit'):
                         nn.init.zeros_(module.bias)
         else:
             if isinstance(module, nn.Linear):
-                trunc_normal_tf_(module.weight, std=.02)
+                nn.init.normal_(module.weight, std=.02)
                 if module.bias is not None:
                     nn.init.zeros_(module.bias)
 
@@ -490,7 +497,6 @@ def no_weight_decay(self):
             k for k, _ in self.named_parameters()
             if any(n in k for n in ["relative_position_bias_table", "rel_pos.mlp"])}
 
-
     @torch.jit.ignore
     def group_matcher(self, coarse=False):
         matcher = dict(
@@ -567,7 +573,6 @@ def gcvit_small(pretrained=False, **kwargs):
     model_kwargs = dict(
         depths=(3, 4, 19, 5),
         num_heads=(3, 6, 12, 24),
-        window_size=(7, 7, 14, 7),
         embed_dim=96,
         mlp_ratio=2,
         layer_scale=1e-5,
@@ -580,7 +585,6 @@ def gcvit_base(pretrained=False, **kwargs):
     model_kwargs = dict(
         depths=(3, 4, 19, 5),
         num_heads=(4, 8, 16, 32),
-        window_size=(7, 7, 14, 7),
         embed_dim=128,
         mlp_ratio=2,
         layer_scale=1e-5,