mirrors/pytorch-image-models
1
0
Fork 0
mirror of https://github.com/huggingface/pytorch-image-models.git synced 2025-06-03 15:01:08 +08:00
Code Issues Projects Releases Wiki Activity

Tweak init for convnext block using maxxvit/coatnext.

Browse Source
This commit is contained in:
Ross Wightman 2022-08-25 15:30:59 -07:00
parent 1d8d6f6072
commit 7c2660576d
1 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
timm/models/maxxvit.py
View File

@ -259,8 +259,6 @@ def _rw_max_cfg(
# - mbconv expansion calculated from input instead of output chs # - mbconv expansion calculated from input instead of output chs
# - mbconv shortcut and final 1x1 conv did not have a bias # - mbconv shortcut and final 1x1 conv did not have a bias
# - mbconv uses silu in timm, not gelu # - mbconv uses silu in timm, not gelu
# - avg pool with kernel_size=2 favoured downsampling (instead of maxpool for coat)
# - default to avg pool for mbconv downsample instead of 1x1 or dw conv
# - expansion in attention block done via output proj, not input proj # - expansion in attention block done via output proj, not input proj
return dict( return dict(
conv_cfg=MaxxVitConvCfg( conv_cfg=MaxxVitConvCfg(
@ -411,12 +409,6 @@ model_cfgs = dict(
rel_pos_dim=384, # was supposed to be 512, woops rel_pos_dim=384, # was supposed to be 512, woops
), ),
), ),
coatnext_nano_rw_224=MaxxVitCfg(
embed_dim=(64, 128, 256, 512),
depths=(3, 4, 6, 3),
stem_width=(32, 64),
**_next_cfg(),
),
coatnet_nano_cc_224=MaxxVitCfg( coatnet_nano_cc_224=MaxxVitCfg(
embed_dim=(64, 128, 256, 512), embed_dim=(64, 128, 256, 512),
depths=(3, 4, 6, 3), depths=(3, 4, 6, 3),
@ -424,6 +416,13 @@ model_cfgs = dict(
block_type=('C', 'C', ('C', 'T'), ('C', 'T')), block_type=('C', 'C', ('C', 'T'), ('C', 'T')),
**_rw_coat_cfg(), **_rw_coat_cfg(),
), ),
coatnext_nano_rw_224=MaxxVitCfg(
embed_dim=(64, 128, 256, 512),
depths=(3, 4, 6, 3),
stem_width=(32, 64),
weight_init='normal',
**_next_cfg(),
),
# Trying to be like the CoAtNet paper configs # Trying to be like the CoAtNet paper configs
coatnet_0_224=MaxxVitCfg( coatnet_0_224=MaxxVitCfg(
@ -498,6 +497,7 @@ model_cfgs = dict(
depths=(1, 2, 3, 1), depths=(1, 2, 3, 1),
block_type=('M',) * 4, block_type=('M',) * 4,
stem_width=(32, 64), stem_width=(32, 64),
weight_init='normal',
**_next_cfg(window_size=8), **_next_cfg(window_size=8),
), ),