The first step to using Manot is setting up the data you'll be working with.
Manot takes ground truth data as input, in the form of a portion of your dataset with ground truth labels, along with your model's predictions. You should have an image path and two labels, one for the ground truth and one for your model's predictions.
For a standard use of Manot, your data should either local or from an Amazon S3 bucket. You can also use Activeloop's Deep Lake through our integration with them.
Below, the detections path is the path to the directory where detections of the model are done with a 5% confidence threshold.
The metadata format must be one of "xyx2y2", "xywh", or "cxcywh".
setup = manot.setup(
data_provider= "data", # it must be "s3" or "local"
arguments={
"name": "setup_example",
"images_path": "/path/to/images",
"ground_truths_path": "/path/to/ground_truths",
"detections_path": "/path/to/detections",
"detections_metadata_format": "xyx2y2",
"classes_txt_path": "/path/to/classes.txt",
"task": 'detection'
}
)
Example
Below is a data setup example for classification tasks.
You can find a full example for classification in our GitHub repository
