'''mask2former_swinlarge_ade20k'''
import os
import copy
from .base_cfg import SEGMENTOR_CFG
from .._base_ import DATASET_CFG_ADE20k_640x640, DATALOADER_CFG_BS16


# deepcopy
SEGMENTOR_CFG = copy.deepcopy(SEGMENTOR_CFG)
# modify dataset config
SEGMENTOR_CFG['dataset'] = DATASET_CFG_ADE20k_640x640.copy()
SEGMENTOR_CFG['dataset']['train']['data_pipelines'][0] = (
    'RandomChoiceResize', {
        'scales': [int(512 * x * 0.1) for x in range(5, 21)], 
        'resize_type': 'ResizeShortestEdge', 
        'max_size': 2048,
    }
)
# modify dataloader config
SEGMENTOR_CFG['dataloader'] = DATALOADER_CFG_BS16.copy()
# modify scheduler config
SEGMENTOR_CFG['scheduler']['max_epochs'] = 130
SEGMENTOR_CFG['scheduler']['min_lr'] = 0.0
SEGMENTOR_CFG['scheduler']['clipgrad_cfg'] = {'max_norm': 0.01, 'norm_type': 2}
# modify other segmentor configs
SEGMENTOR_CFG['num_classes'] = 150
SEGMENTOR_CFG['backbone'] = {
    'type': 'SwinTransformer', 'structure_type': 'swin_large_patch4_window12_384_22k', 'pretrained': True, 
    'selected_indices': (0, 1, 2, 3), 'norm_cfg': {'type': 'LayerNorm'},
    'pretrain_img_size': 384, 'in_channels': 3, 'embed_dims': 192, 'patch_size': 4, 'window_size': 12, 'mlp_ratio': 4,
    'depths': [2, 2, 18, 2], 'num_heads': [6, 12, 24, 48], 'qkv_bias': True, 'qk_scale': None, 'patch_norm': True,
    'drop_rate': 0., 'attn_drop_rate': 0., 'drop_path_rate': 0.3, 'use_abs_pos_embed': False,
}
SEGMENTOR_CFG['head']['pixel_decoder']['input_shape']['in_channels'] = [192, 384, 768, 1536]
SEGMENTOR_CFG['work_dir'] = os.path.split(__file__)[-1].split('.')[0]
SEGMENTOR_CFG['logger_handle_cfg']['logfilepath'] = os.path.join(SEGMENTOR_CFG['work_dir'], f"{os.path.split(__file__)[-1].split('.')[0]}.log")
# modify training tricks in scheduler config
SEGMENTOR_CFG['scheduler']['optimizer']['params_rules'] = {
    'base_setting': dict(norm_wd_multiplier=0.0),
    'backbone_net': dict(lr_multiplier=0.1, wd_multiplier=1.0),
    'backbone_net.patch_embed.norm': dict(lr_multiplier=0.1, wd_multiplier=0.0),
    'backbone_net.norm': dict(lr_multiplier=0.1, wd_multiplier=0.0),
    'absolute_pos_embed': dict(lr_multiplier=0.1, wd_multiplier=0.0),
    'relative_position_bias_table': dict(lr_multiplier=0.1, wd_multiplier=0.0),
    'query_embed': dict(lr_multiplier=1.0, wd_multiplier=0.0),
    'query_feat': dict(lr_multiplier=1.0, wd_multiplier=0.0),
    'level_embed': dict(lr_multiplier=1.0, wd_multiplier=0.0),
}
for stage_id, num_blocks in enumerate(SEGMENTOR_CFG['backbone']['depths']):
    for block_id in range(num_blocks):
        SEGMENTOR_CFG['scheduler']['optimizer']['params_rules'].update({
            f'backbone_net.stages.{stage_id}.blocks.{block_id}.norm': dict(lr_multiplier=0.1, wd_multiplier=0.0)
        })
for stage_id in range(len(SEGMENTOR_CFG['backbone']['depths']) - 1):
    SEGMENTOR_CFG['scheduler']['optimizer']['params_rules'].update({
        f'backbone_net.stages.{stage_id}.downsample.norm': dict(lr_multiplier=0.1, wd_multiplier=0.0)
    })