mmselfsup/configs/selfsup/_base_/models/beit_vit-base-p16.py
RenQin 7a7b048f23
[Feature]: Add BEiT Support (#425)
* [Feature]: Add BEiT Support

* [Fix]: fix bugs after update

* [Fix]: fix bugs in backbone

* [Refactor]: refactor config

* [Feature]: Support BEiTv2

* [Fix]: Fix UT

* [Fix]: rename some configs

* [Fix]: fix beitv2neck

* [Refactor]: refactor beitv2

* [Fix]: fix lint

* refactor configs

* refactor beitv2

* update configs

* add dalle target generator

* refactor for beitv1

* refactor rel_pos_bias of beit

* update configs

* update configs

* update v1 configs

* update v2 configs

* refactoe layer decay

* update unittest

* fix lint

* fix ut

* add docstrings

* rename

* fix lint

* add beit model and log links

* fix lint

* update according to review

* update

* update

* update LearningRateDecayOptimWrapperConstructor
related configs

* update init and backbone

* update neck and vqkd

* refactor neck

* fix lint

* add some comments

* fix typo

Co-authored-by: 任琴 <PJLAB\renqin@shai14001114l.pjlab.org>
Co-authored-by: fangyixiao18 <fangyx18@hotmail.com>
2022-12-06 16:40:05 +08:00

29 lines
865 B
Python

# model settings
model = dict(
type='BEiT',
backbone=dict(
type='BEiTViT',
arch='base',
patch_size=16,
drop_path_rate=0.1,
final_norm=True,
layer_scale_init_value=0.1,
init_cfg=[
dict(type='TruncNormal', std=0.02, layer='Linear'),
dict(type='TruncNormal', std=0.02, layer='Conv2d'),
dict(type='Constant', layer='LayerNorm', val=1.0, bias=0.0)
]),
neck=None,
head=dict(
type='BEiTV1Head',
embed_dims=768,
num_embed=8192,
loss=dict(type='BEiTLoss')),
target_generator=dict(
type='DALL-E',
init_cfg=dict(
type='Pretrained',
checkpoint= # noqa: E251
'https://download.openmmlab.com/mmselfsup/1.x/target_generator_ckpt/dalle_encoder.pth', # noqa: E501
)))