Clarifai for now only tags ".jpg", ".jpeg" and ".png" images. So limit yourself to those images. The script takes a csv file that includes links to image files (see sample input file). And the script appends the following new columns to the csv (see sample output file):
-
clarifai_status:- None: URLs haven't tagged by Clarifai (include URLs was filtered out)
- OK: Tagged by Clarifai with OK status
- ALL_ERROR => Failed to tag by Clarifai, if there is video URL in the batch.
- CLIENT_ERROR => Failed to tag by Clarifai
- ERROR => if there is Error in the script
In case the status is "ALL_ERROR", we can reset these errors manually (just delete the status) and re-run the script with a smaller batch size. (-b option in
clarify_tag.py) -
tags: Tags returned by clarifai -
probs: Probability of each tag being applicable, returned by clarifai -
predicted: Predicted gender for the person(s) in the image
To begin, install the requirements. In particular, the script depends on pandas and clarifai API client.
Before running the script, please set app_id and app_secret in clarifai.cfg or make sure the environment variables CLARIFAI_APP_ID and CLARIFAI_APP_SECRET are set.
usage: clarifai_tag.py [-h] [--config CONFIG] [-c COUNT] [-b BATCH]
[-o OUTPUT] [--no-header]
input
Clarifai Tag API
positional arguments:\github\clarifai_gender\dev\scripts\get_clarifai_tags
input Input file name
optional arguments:
-h, --help show this help message and exit
--config CONFIG Configuration file
-c COUNT, --count COUNT
Number of valid tag for each item
-b BATCH, --batch BATCH
Specify batch size
-o OUTPUT, --output OUTPUT
Output CSV file name
--no-header Output without header at the first row
To get at least 20 tags per image:
python clarifai_tag.py sample_in.csv -c 20
Default name of the output file is output-img-tag.csv.