Detecting more than 100 objects in a single image using retinanet

[DKosatov]

Hello, could you explain if it is possible to use retinanet to find more than 100 objects in one image?

My images may contain more than 100 objects of the same type.
I'm setting max_detections=400 for MultiClassNonMaxSuppression.

keras_cv.layers.MultiClassNonMaxSuppression(
    bounding_box_format="xyxy",
    from_logits=True,
    iou_threshold=0.2,
    confidence_threshold=0.6,
    max_detections=400
)

But

y_pred = mt.predict(image_batch)
print(y_pred['confidence'][0])

It always contains no more than 100 predicted objects and 300 (-1.).
Please tell me if there is a way to fix this

[DKosatov]

Since the objects in the image were of the same class, it was necessary to add
max_detections_per_class=400
After that, everything works correctly.