[Merge After 0.6.1] Adding support for str/list input for classification #2541
Conversation
yongxinw
changed the title
|
Job PR-2541-4487c0d is done. |
| if contains_valid_images(data): | ||
| data_dict = {"image": data} |
There was a problem hiding this comment.
Is it necessary to check if a list of image paths and use header image? Whether image list or not doesn't affect creating the dataframe. pandas creates a dataframe with header 0. I don't think header image is required in later data processing and model forward.
There was a problem hiding this comment.
let me double check. I think I saw at some point the code actually uses the header image
| @@ -515,3 +526,38 @@ def infer_dtypes_by_model_names(model_config: DictConfig): | |||
| fallback_dtype = list(allowable_dtypes)[0] | |||
|
|
|||
| return allowable_dtypes, fallback_dtype | |||
|
|
|||
|
|
|||
| def contains_valid_images(data: Union[str, list], sample_n: Optional[int] = 50) -> bool: | |||
There was a problem hiding this comment.
How about reusing is_imagepath_column() in infer_types.py?
There was a problem hiding this comment.
the is_imagepath_column checks on a df.Series. This test was for array of images. That being said, let me check if the image header is needed. If not, I can convert the image array to pd.Series and re-use is_imagepath_column().
| elif isinstance(data, str): | ||
| data = load_pd.load(data) | ||
| if contains_valid_images(data): | ||
| data_dict = {"image": [data]} |
There was a problem hiding this comment.
Similarly, header image seems unnecessary.
FANGAreNotGnu
changed the title
Issue #, if available:
Description of changes: Adding support for str/list input for classification. Unit Test passed.
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.