Annotation file support #32
Conversation
- pycocotools and pyyaml requirements added. - I/O utils for json/yaml added.
…ons of Annotations class. required updates also made to load_from_coco and persist_as_yolo methods
There was a problem hiding this comment.
@gauravparajuli Thank you for this amazing work. I've reviewed the code, and have few points on the implementation. We can proceed with the merge after resolving those issues.
| all_files.extend(glob.glob(f"{directory}/*{ext}")) | ||
| return Counter(file.split('.')[-1] for file in all_files) | ||
|
|
||
| sample_yolo_dataset_path = str(os.path.join('tests', 'pybboxes', 'annotations', 'testing_data_yolo')) |
There was a problem hiding this comment.
Can we upload these fixtures to a Huggingface ? and the test can run on the fly without the local dependency of the dataset (if exists, use, else, download) (huggingface python sdk already does this caching thing for you, then we will need only path-names or urls here.) I think it would be a nicer way of adding these kind of heavy fixtures.
pybboxes/annotations/base.py
Outdated
| else: | ||
| self._objects[image_name].append(annotation) | ||
|
|
||
| def persist_as_yolo(self, export_dir: str): |
There was a problem hiding this comment.
can we rename the methods as write_as_... or save_as_... I think it would be better if we stick to the global naming convention with the other python packages. I'd suggest either read/write or load/save, both are fine I think.
There was a problem hiding this comment.
hello, I have made the revisions that you requested for
|
Thanks @gauravparajuli , can you reformat the codebase as instructed at README, to run the tests. if your local codebase is not added to the python -m tests.run_code_style format |
Hi, I have formatted the code |
|
Hi @gauravparajuli, can you please recheck you code formatting and apply it again ? It shows error on the code formatting still (see this). You can locally check if the codebase is correctly formatted as follows: python -m tests.run_code_style check |
gauravparajuli
force-pushed
the
annotation-file-support
branch
from
f0c9020 to
904558c
Compare
hey,
if you want a high level overview on how to use this new feature:
https://github.com/gauravparajuli/pybboxes/tree/annotation-file-support?tab=readme-ov-file#annotation-file-conversion
I included a small coco dataset (196 files) with about 8 megabytes size in the testing directory to check the conversions during unit tests.
currently, also, attempts to either load or write as albumentations and fiftyone raises a not implemented error.
please let me know what you think of this.