Open source network:

https://github.com/NVlabs/SegFormer

When you train segformer, you report an error:

“EncoderDecoder: ‘mit_ b1 is not in the backbone registry'”

The direct cause of this exception is:

Find MIT in the dictionary_ B1, throw exception if not found:

    obj_type = args.pop('type')
    if isinstance(obj_type, str):
        obj_cls = registry.get(obj_type)
        if obj_cls is None:
            print(obj_type)
            raise KeyError(
                f'{obj_type} is not in the {registry.name} registry')

mit_ B1 is a class,

@BACKBONES.register_module()
class mit_b1(MixVisionTransformer):
    def __init__(self, **kwargs):
        super(mit_b1, self).__init__(
            patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[4, 4, 4, 4],
            qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[2, 2, 2, 2], sr_ratios=[8, 4, 2, 1],
            drop_rate=0.0, drop_path_rate=0.1)

Self simulation exception:

Method 1, put the class MIT_ If B1 is commented out, the above exception will be reported, MIT_ B1 not registered

Method 2:

In mmseg/Models/builder.py

Add test code:

bbb= BACKBONES.get(‘mit_ b2’)

print(“bbb”,bbb)

Complete as follows:

import warnings

from mmcv.utils import Registry, build_from_cfg
from torch import nn

BACKBONES = Registry('backbone')
NECKS = Registry('neck')
HEADS = Registry('head')
LOSSES = Registry('loss')
SEGMENTORS = Registry('segmentor')

bbb= BACKBONES.get('mit_b2')

print("bbb",bbb)

The result BBB is empty, MIT_ B1 not registered

Method 3:

# from mmseg.models import BACKBONES
from mmseg.models.builder import BACKBONES

bbb= BACKBONES.get('mit_b1')

print("bbb",bbb)

The result BBB is empty, MIT_ B1 not registered

Find and quote MIT again_ B1 file

mix_transformer.py

You can register successfully. The code is as follows:

Create a registry under the root directory_ Demo.py, the test code is as follows:

# from mmseg.models import BACKBONES
from mmseg.models.backbones import mix_transformer
from mmseg.models.builder import BACKBONES

# from .mix_transformer import *


bbb= BACKBONES.get('mit_b1')

print("bbb2",bbb)

bbb2 < class ‘mmseg.models.backbones.mix_ transformer.mit_ b1’>

Normal test method:

In the root directory,

Create a registry_ Demo.py, the test code is as follows:

If the registration is successful, bbb2 is not empty. If the registration is not successful, bbb2 is empty.

from mmseg.models import BACKBONES

bbb= BACKBONES.get('mit_b1')

print("bbb2",bbb)