Skip to content

Latest commit

 

History

History
260 lines (191 loc) · 7.52 KB

add_models.md

File metadata and controls

260 lines (191 loc) · 7.52 KB
Raw

Add New Modules

Develop new components

We can customize all the components introduced at the model documentation, such as backbone, head, loss function and data preprocessor.

Add new backbones

Here we show how to develop a new backbone with an example of MobileNet.

  1. Create a new file mmseg/models/backbones/mobilenet.py.

    import torch.nn as nn

from mmseg.registry import MODELS


@MODELS.register_module()
class MobileNet(nn.Module):

    def __init__(self, arg1, arg2):
        pass

    def forward(self, x):  # should return a tuple
        pass

    def init_weights(self, pretrained=None):
        pass

  2. Import the module in mmseg/models/backbones/__init__.py.

    from .mobilenet import MobileNet

  3. Use it in your config file.

    model = dict(
    ...
    backbone=dict(
        type='MobileNet',
        arg1=xxx,
        arg2=xxx),
    ...

Add new heads

In MMSegmentation, we provide a BaseDecodeHead for developing all segmentation heads. All newly implemented decode heads should be derived from it. Here we show how to develop a new head with the example of PSPNet as the following.

First, add a new decode head in mmseg/models/decode_heads/psp_head.py. PSPNet implements a decode head for segmentation decode. To implement a decode head, we need to implement three functions of the new module as the following.

from mmseg.registry import MODELS

@MODELS.register_module()
class PSPHead(BaseDecodeHead):

    def __init__(self, pool_scales=(1, 2, 3, 6), **kwargs):
        super(PSPHead, self).__init__(**kwargs)

    def init_weights(self):
        pass

    def forward(self, inputs):
        pass

Next, the users need to add the module in the mmseg/models/decode_heads/__init__.py, thus the corresponding registry could find and load them.

To config file of PSPNet is as the following

norm_cfg = dict(type='SyncBN', requires_grad=True)
model = dict(
    type='EncoderDecoder',
    pretrained='pretrain_model/resnet50_v1c_trick-2cccc1ad.pth',
    backbone=dict(
        type='ResNetV1c',
        depth=50,
        num_stages=4,
        out_indices=(0, 1, 2, 3),
        dilations=(1, 1, 2, 4),
        strides=(1, 2, 1, 1),
        norm_cfg=norm_cfg,
        norm_eval=False,
        style='pytorch',
        contract_dilation=True),
    decode_head=dict(
        type='PSPHead',
        in_channels=2048,
        in_index=3,
        channels=512,
        pool_scales=(1, 2, 3, 6),
        dropout_ratio=0.1,
        num_classes=19,
        norm_cfg=norm_cfg,
        align_corners=False,
        loss_decode=dict(
            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)))

Add new loss

Assume you want to add a new loss as MyLoss for segmentation decode. To add a new loss function, the users need to implement it in mmseg/models/losses/my_loss.py. The decorator weighted_loss enables the loss to be weighted for each element.

import torch
import torch.nn as nn

from mmseg.registry import MODELS
from .utils import weighted_loss

@weighted_loss
def my_loss(pred, target):
    assert pred.size() == target.size() and target.numel() > 0
    loss = torch.abs(pred - target)
    return loss

@MODELS.register_module()
class MyLoss(nn.Module):

    def __init__(self, reduction='mean', loss_weight=1.0):
        super(MyLoss, self).__init__()
        self.reduction = reduction
        self.loss_weight = loss_weight

    def forward(self,
                pred,
                target,
                weight=None,
                avg_factor=None,
                reduction_override=None):
        assert reduction_override in (None, 'none', 'mean', 'sum')
        reduction = (
            reduction_override if reduction_override else self.reduction)
        loss = self.loss_weight * my_loss(
            pred, target, weight, reduction=reduction, avg_factor=avg_factor)
        return loss

Then the users need to add it in the mmseg/models/losses/__init__.py.

from .my_loss import MyLoss, my_loss

To use it, modify the loss_xxx field. Then you need to modify the loss_decode field in the head. loss_weight could be used to balance multiple losses.

loss_decode=dict(type='MyLoss', loss_weight=1.0))

Add new data preprocessor

In MMSegmentation 1.x versions, we use SegDataPreProcessor to copy data to the target device and preprocess the data into the model input format as default. Here we show how to develop a new data preprocessor.

  1. Create a new file mmseg/models/my_datapreprocessor.py.

    from mmengine.model import BaseDataPreprocessor

from mmseg.registry import MODELS

@MODELS.register_module()
class MyDataPreProcessor(BaseDataPreprocessor):
    def __init__(self, **kwargs):
        super().__init__(**kwargs)

    def forward(self, data: dict, training: bool=False) -> Dict[str, Any]:
        # TODO Define the logic for data pre-processing in the forward method
        pass

  2. Import your data preprocessor in mmseg/models/__init__.py

    from .my_datapreprocessor import MyDataPreProcessor

  3. Use it in your config file.

    model = dict(
    data_preprocessor=dict(type='MyDataPreProcessor)
    ...
)

Develop new segmentors

The segmentor is an algorithmic architecture in which users can customize their algorithms by adding customized components and defining the logic of algorithm execution. Please refer to the model document for more details.

Since the BaseSegmentor in MMSegmentation unifies three modes for a forward process, to develop a new segmentor, users need to overwrite loss, predict and _forward methods corresponding to the loss, predict and tensor modes.

Here we show how to develop a new segmentor.

  1. Create a new file mmseg/models/segmentors/my_segmentor.py.

     from typing import Dict, Optional, Union

 import torch

 from mmseg.registry import MODELS
 from mmseg.models import BaseSegmentor

 @MODELS.register_module()
 class MySegmentor(BaseSegmentor):
     def __init__(self, **kwargs):
         super().__init__(**kwargs)
         # TODO users should build components of the network here

     def loss(self, inputs: Tensor, data_samples: SampleList) -> dict:
         """Calculate losses from a batch of inputs and data samples."""
         pass

     def predict(self, inputs: Tensor, data_samples: OptSampleList=None) -> SampleList:
         """Predict results from a batch of inputs and data samples with post-
         processing."""
         pass

    def _forward(self,
              inputs: Tensor,
              data_samples: OptSampleList = None) -> Tuple[List[Tensor]]:
         """Network forward process.

         Usually includes backbone, neck and head forward without any post-
         processing.
         """
         pass

  2. Import your segmentor in mmseg/models/segmentors/__init__.py.

    from .my_segmentor import MySegmentor

  3. Use it in your config file.

    model = dict(
    type='MySegmentor'
    ...
)