[Doc] Add dataflow document

[xiexinch]

Motivation

Add dataflow doc

Modification

• docs/en/advanced_guides/data_flow.md

[codecov]

Codecov Report

Base: 83.33% // Head: 83.33% // No change to project coverage 👍

Coverage data is based on head (6293e07) compared to base (6cb64e3).
Patch has no changes to coverable lines.

❗ Current head 6293e07 differs from pull request most recent head ec4dd03. Consider uploading reports for the commit ec4dd03 to get more accurate results

Additional details and impacted files

@@           Coverage Diff            @@
##           dev-1.x    #2403   +/-   ##
========================================
  Coverage    83.33%   83.33%           
========================================
  Files          143      143           
  Lines         8127     8127           
  Branches      1211     1211           
========================================
  Hits          6773     6773           
  Misses        1165     1165           
  Partials       189      189           

Flag	Coverage Δ
unittests	83.33% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[MengzhangLI]

Add specific http link of models.md.

[MengzhangLI]

Suggested change

	In MMSegmentation, the method [loss_by_feat](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/mmseg/models/decode_heads/decode_head.py#L291) of `decode_head` is a unify interface used to compute loss.
	In MMSegmentation, the method [loss_by_feat](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/mmseg/models/decode_heads/decode_head.py#L291) of `decode_head` is an unified interface used to compute loss.

[MeowZheng]

It should emphasize that Runner controls the dataflow, and add examples of train_cfg, test_cfg and val_cfg in mmseg. might also attach the link for runner design documentation https://github.com/open-mmlab/mmengine/blob/main/docs/en/design/runner.md

it is necessary to explain the train_step, val_step or test_step work for each iteration for training and testing and add links https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md#train_step, https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md#val_step and https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md#test_step

[MeowZheng]

Suggested change

	## Format convention
	## Dataflow convention in MMSegmentation

[MeowZheng]

Suggested change

	### Model to Evaluator
	### Model output
	#### To Evaluator
	#### Optim Wrapper
	####

[MeowZheng]

we can add some explanation about the control flow of runner when training and testing, like call train_step: model.forward-> evaluator, call test_step: model.forward -> optim wrapper refer, and attach the link https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md

and highlight the input and output of model and other modules

[MeowZheng]

Suggested change

Without any modifications, the return value of PackSegInputs is usually a `dict` and has only two keys, `inputs` and `data_samples`. The following pseudo-code shows the data types of the data loader output, `inputs` is the list of input tensors to the model and `data_samples` contains a list of input images' meta information and corresponding ground truth.

Without any modifications, the return value of PackSegInputs is usually a `dict` and has only two keys, `inputs` and `data_samples`. The following pseudo-code shows the data types of the data loader output in mmseg, which is a batch of fetched data samples from the dataset, and data loader packs them into a dictionary of the list. `inputs` is the list of input tensors to the model and `data_samples` contains a list of input images' meta information and corresponding ground truth.

[MeowZheng]

Suggested change

	Though drawn separately in the diagram [above](#overview-of-dataflow), data_preprocessor is a part of the model and thus can be found in [Model tutorial](https://mmsegmentation.readthedocs.io/en/dev-1.x/advanced_guides/models.html) at Seg DataPreprocessor chapter.
	Though drawn separately in the diagram [above](#overview-of-dataflow), data_preprocessor is a part of the model and thus can be found in [Model tutorial](https://mmsegmentation.readthedocs.io/en/dev-1.x/advanced_guides/models.html) at data preprocessor chapter.

[MeowZheng]

Suggested change

The return value of Data Preprocessor is a dict, contains `inputs` and `data_samples`, `inputs` would be transferred to GPU and some additional metainfo like `pad_shape` and `padding_size` would be added to the `data_samples`. When transfer to the network, the dict would be unpacked to two values. The following pseudo-codes show the return value of data preprocessor and the input values of model.

The return value of data preprocessor is a dictionary, containing `inputs` and `data_samples`, `inputs` is batched images, a 4D tensor, and some additional meta info used in data preprocesses would be added to the `data_samples`. When transferred to the network, the dictionary would be unpacked to two values. The following pseudo-codes show the return value of the data preprocessor and the input values of model.

[MeowZheng]

Suggested change

	```python
	dict(
	inputs=torch.Tensor,
	data_samples=Optional[List[SegDataSample], None]
	)
	```
	```python
	dict(
	inputs=torch.Tensor,
	data_samples=List[SegDataSample]
	)

I think this tutorial is about data flow under runner control, which contains gt always

[MeowZheng]

Suggested change

In the above diagram, the red line indicates the [train_step](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md#train_step). At each training iteration, dataloader loads images from storage and transfer to data preprocessor, data preprocessor would put images to GPU device and stack data to batch, then model accept the batch data as inputs, finally the outputs of the model would be sent to optimizer. The blue line indicates [val_step](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md#val_step) and [test_step](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md#test_step). The dataflow of these two process is similar to the `train_step` except the outputs of model, since model parameters are freezed when doing evaluation, the model output would be transferred to [Evaluator](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/evaluation.md#ioumetric) to compute metrics.

In the above diagram, the red line indicates the [train_step](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md#train_step). At each training iteration, dataloader loads images from storage and transfer to data preprocessor, data preprocessor would put images to the specific device and stack data to batch, then model accepts the batch data as inputs, finally the outputs of the model would be sent to optimizer. The blue line indicates [val_step](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md#val_step) and [test_step](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md#test_step). The dataflow of these two process is similar to the `train_step` except the outputs of model, since model parameters are freezed when doing evaluation, the model output would be transferred to [Evaluator](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/evaluation.md#ioumetric) to compute metrics.

[MeowZheng]

Suggested change

	#### To Evaluator
	As [model tutorial](https://github.com/open-mmlab/mmsegmentation/blob/dev-1.x/docs/en/advanced_guides/models.md#forward) mentioned 3 kinds of mode forward with 3 kinds of output. `train_step`and `test_step`(or `val_step`) correspond to `'loss'` and `'predict'` respectively.

[MeowZheng]

Suggested change

	At the evaluation procedure, the inference results would be transferred to `Evaluator`. You might read the [evaluation document](https://mmsegmentation.readthedocs.io/en/dev-1.x/advanced_guides/evaluation.html) for more information about `Evaluator`.
	In `test_step` or `val_step`, the inference results would be transferred to `Evaluator`. You might read the [evaluation document](https://mmsegmentation.readthedocs.io/en/dev-1.x/advanced_guides/evaluation.html) for more information about `Evaluator`.

[MeowZheng]

Suggested change

#### Optim Wrapper