xcenternet icon indicating copy to clipboard operation
xcenternet copied to clipboard

Adding custom metric not working

Open Sara-Samer opened this issue 1 year ago • 0 comments

I'm trying to add a custom metric to calculate mean AP while training but it's not working the model is not printing it or calling it's update state


import tensorflow as tf
from xcenternet.model.constants import MAX_OBJECTS, IOU_THRESHOD, MAP_SCORE_THRESHOLD, LABELS
from xcenternet.model.evaluation.mean_average_precision import MAP

class MeanAP(tf.keras.metrics.Metric):
    def __init__(self, name="custom_mean_ap", **kwargs):
        super(MeanAP, self).__init__(name=name, **kwargs)
        self.iou_threshold = IOU_THRESHOD
        self.score_threshold = MAP_SCORE_THRESHOLD
        self.mean_average_precision = MAP(LABELS, iou_threshold=self.iou_threshold, score_threshold=self.score_threshold)
        self.map = self.add_weight(name="ctp", initializer="zeros")
        self.batch_count = 0

    def update_state(self, training_data, predictions):
        mask, bboxes, labels = training_data["mask"], training_data["bboxes"], training_data["labels"]
        decoded = self.decode(predictions, relative=False, k=MAX_OBJECTS)
        self.mean_average_precision.update_state_batch(decoded, bboxes, labels, mask)
        result = self.mean_average_precision.result()
        self.map.assign_add(result*len(training_data["labels"]))
        self.batch_count += len(training_data["labels"])
    
    def result(self):
        return self.map/self.batch_count

    def reset_state(self):
        self.batch_count = 0
        self.map.assign(0.0)

I added the metric to compile

    model.compile(optimizer=optimizer, loss=model.get_loss_funcs(), metrics=[MeanAP()])

Sara-Samer avatar Dec 04 '22 13:12 Sara-Samer