Hello I'm having issues with PyQt5 on my current system and the push buttons for f

<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/us

Provide example of use with command line interface about review_object_detection_metrics HOT 2 CLOSED

rafaelpadilla commented on May 29, 2024

Provide example of use with command line interface

from review_object_detection_metrics.

Comments (2)

potipot commented on May 29, 2024 4

I managed to run it by inspecting the code. Do you plan to convert it to a pip library?

Here is my code, maybe somebody will find it useful.

def raw_to_odm(preds, records):
# preds and records are in custom format of lists of BBox from icevision: https://github.com/airctic/icevision
    groundtruth_bbs = []
    detected_bbs = []
    for pred, record in zip(preds, records):
        image_name = record['filepath'].name
        for bbox, label in zip(record['bboxes'], record['labels']):
            class_id=label
            coordinates = bbox.xywh
            bb = BoundingBox(image_name=image_name, class_id=class_id, coordinates=coordinates)
            groundtruth_bbs.append(bb)

        for score, bbox, label in zip(pred['scores'], pred['bboxes'], pred['labels']):
            class_id=label
            coordinates = bbox.xywh
            bb = BoundingBox(image_name=image_name, class_id=class_id, coordinates=coordinates, 
                             bb_type=BBType.DETECTED, confidence=score)
            detected_bbs.append(bb)
    return groundtruth_bbs, detected_bbs

get_coco_summary(groundtruth_bbs, detected_bbs)

output:

{'AP': 0.1096970274832816,
 'AP50': 0.21066336753733456,
 'AP75': 0.09695353970144305,
 'APsmall': 0.004913206881231863,
 'APmedium': 0.19432812877332575,
 'APlarge': 0.2652736296925409,
 'AR1': 0.15589935094323745,
 'AR10': 0.2210287542595406,
 'AR100': 0.23329693548873054,
 'ARsmall': 0.06724119108517042,
 'ARmedium': 0.32722893724006674,
 'ARlarge': 0.3624383693902665}

from review_object_detection_metrics.

rafaelpadilla commented on May 29, 2024

@potipot

I'm glad you got it working. We are planning to make the evaluation accessible with a CLI.

We do not have plans to make it a pip library yet. First, it would be good to collect more feedback from users before releasing via pip.

Rafael

from review_object_detection_metrics.

Recommend Projects