nlpaueb / biocaption Goto Github PK

Split dataset percentage is now hardcoded.
Change it to a user-changeable parameter.

The models accept a list of images and not only one image.

What's the form that a dataset needs to have in order to be processed by the code?

Add python logger for debugging.

Usage:

from bioCaption.data.downloads import DownloadData
downloads = DownloadData()
downloads.download_peir_gross()

Implement logic where if embeddings do not exist then download them.

Users can define where to download the datasets.

User has the possibility to give their own dataset.

Hello,

I think this line:
evaluation = Evaluation('iu_xray/test_images.tsv', 'results/most_frequent_word_results.tsv')
in README.md should be changed to this:
evaluation = Evaluation('results/most_frequent_word_results.tsv', 'iu_xray/test_images.tsv')

according to:
https://github.com/nlpaueb/dc/blob/211c9c70f145e71082a45376dd2cab846441023e/dc/evaluation/evaluation.py#L20

Edit: I have now made a pull request that solves this issue and fixes the README file

I just started using this package so excuse me if I'm ever mistaken in any issues that I raise.
Using the package, I get a lot of errors about function names. Browsing through the files, I saw that the actual call names of those functions have a leading underscore. I made this issue yesterday with 1 line of code but now I'm editing it to reference more:

https://github.com/nlpaueb/dc/blob/211c9c70f145e71082a45376dd2cab846441023e/dc/data/downloads.py#L14
(_get_logger might not need the leading underscore in dc/configuration.py, since it doesn't seem like a 'private' function from what I saw, so this line of code may be correct actually. Note that if you delete the leading underscore in dc/configuration.py, you'll also have to delete it in lines 7 and 9 of dc/evaluation/evaluation.py)

https://github.com/nlpaueb/dc/blob/211c9c70f145e71082a45376dd2cab846441023e/dc/evaluation/evaluation.py#L59

https://github.com/nlpaueb/dc/blob/211c9c70f145e71082a45376dd2cab846441023e/dc/evaluation/evaluation.py#L102

https://github.com/nlpaueb/dc/blob/211c9c70f145e71082a45376dd2cab846441023e/dc/evaluation/evaluation.py#L63

https://github.com/nlpaueb/dc/blob/211c9c70f145e71082a45376dd2cab846441023e/dc/evaluation/evaluation.py#L64

https://github.com/nlpaueb/dc/blob/211c9c70f145e71082a45376dd2cab846441023e/dc/evaluation/evaluation.py#L106

https://github.com/nlpaueb/dc/blob/211c9c70f145e71082a45376dd2cab846441023e/dc/evaluation/evaluation.py#L107

Baselines, Models for many pictures to many tags/captions and models for one image to one caption.

Tsv to json

Set up a set of rules for model contributors.

Users can choose to compute additional statistics while downloading a dataset.
These statistics contain:

print("Found", len(reports_with_no_image), "reports with no associated image")
print("Found", len(reports_with_empty_sections), "reports with empty Impression and Findings sections")
print("Found", len(reports_with_no_impression), "reports with no Impression section")
print("Found", len(reports_with_no_findings), "reports with no Findings section")
print("Collected", len(images_captions), "image-caption pairs")

Add doctrings for the documentation generation.

Installation failed with the following message:

ERROR: No matching distribution found for pycocoevalcap (from bioCaption)
Probably this can be solved with:
pip install git+https://github.com/salaniz/pycocoevalcap
as suggested by:
https://github.com/jialinwu17/pycocoevalcap

Add new functionality implemented in full_reports branch from bio_images_caption