multimodal label-studio export reader & doc #2615
Conversation
|
Job PR-2615-51ee391 is done. |
There was a problem hiding this comment.
Thanks for the PR, left a few comments.
@sxjscience Please let us know what else is missing from current PR, thank you.
| @@ -0,0 +1,173 @@ | |||
| # Label-Studio Export file reader | |||
|
|
|||
There was a problem hiding this comment.
Please use _ in file name, e.g., LabelStudio_export_file_reader.md
|
|
||
| params: | ||
| - path: str: the path of the exported file | ||
| - data_columns: list[str]: the key/column names of the data |
There was a problem hiding this comment.
Can you give a specific example of what data_columns and label_columns look like?
There was a problem hiding this comment.
I will update the specific usage of this two params in the docs, and address the doc link in the code for the next commit. Is it acceptable or should I provide examples as well in the source code as a comment?
| if len(label_studio_json) == 0: | ||
| raise ValueError("ERROR: empty export file") | ||
|
|
||
| if "annotations" in label_studio_json[0]: |
There was a problem hiding this comment.
Why label_studio_json[0]? How many elements are there in label_studio_json? Usually we want to avoid hard coded stuff, like index 0, unless there is good reason.
There was a problem hiding this comment.
For exporting annotations in label-studio, "JSON" and "JSON-MIN" are two different options with the same file extension ".json" that contains a list of annotation dicts. (their examples can be seen in https://labelstud.io/guide/export.html#Label-Studio-JSON-format-of-annotated-tasks and https://labelstud.io/guide/export.html#JSON-MIN, basically JSON-MIN is the simplified version of JSON).
This line of code here is to check if the export file is from "JSON" or "JSON-MIN". Currently I just use a simple check on whether one of the elements (an annotation) has a key "annotations" that exists in JSON but not in JSON-MIN. Still I'm finding better ways to distinguish them and any suggestion is welcomed.
|
|
||
| else: | ||
| split_lst = s.split("/") | ||
| if split_lst[2] == "local-files": |
There was a problem hiding this comment.
Again, how many elements are there in split_lst and what does split_lst[2] represent?
| from autogluon.multimodal.utils import LabelStudioReader | ||
|
|
||
| # initialize LabelStudioReader with default localhost host | ||
| ls=LabelStudioReader() |
There was a problem hiding this comment.
Need extra whitespaces, e.g., ls = LabelStudioReader. There are also many instances like this below, please add whitespaces accordingly.
|
|
||
|
|
||
| """ | ||
| Usage: |
There was a problem hiding this comment.
Try to put the docstring inside “LabelStudioReader” class.
|
Actually, should we rename |
|
Job PR-2615-f886e8e is done. |
There was a problem hiding this comment.
LGTM, we can fill in more details later. @zhiqiangdon @cheungdaven @FANGAreNotGnu @suzhoum @yongxinw Please help to review this PR. This PR enables AG to train on labeled data using LabelStudio, which is a quite useful feature.
bryanyzhu
requested review from
zhiqiangdon,
cheungdaven,
FANGAreNotGnu,
suzhoum and
yongxinw
This tool is to help user to transform the exported label annotation data from a data labeling platform Label-Studio (https://labelstud.io/) and generate the pandas Dataframe for Autogluon multimodal input. In this way use can build up a labelstudio-autogluon workflow, label the data through Label-Studio and then feed the data to Autogluon with a few lines of simple code to adjust the data.
So far there are 3 task template available, including image-classification (image), named entity recognition(text) and user-customized template. Other templates are WIP.
A documentation for this feature is attached to this PR.
Description of changes:
from_labelstudio.pyto autogluon/multimodal/src/autogluon/multimodal/utilslabel-studio-export-readerto autogluon/examples/autommBy submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.