Allen Goodman

And send a PR! It’ll be easier to discuss. 😎

My simple NumPy implementation: ```Python numpy.mean(numpy.cumsum(y_true[numpy.argsort(~y_pred)]) / numpy.cumsum(y_true[numpy.argsort(~y_true)])) ``` Does that look right?

The ImageSegmentationGenerator should be an easier task. The dictionary should look similar to the existing ObjectDetectionGenerator dictionary except each box includes a `mask` key that has the mask pathname, e.g....

TensorFlow implementation: https://www.tensorflow.org/versions/r0.11/api_docs/python/image/cropping#crop_and_resize

It works if I explicitly disable the learning phase: ```Python keras.backend.set_learning_phase(False) model = keras_rcnn.models.RPN(image, classes=len(classes) + 1) model.compile("adam") x, _ = generator.next() prediction = model.predict(x) ``` I suppose this confirms...

@hgaiser Yep! 👍 (I just reviewed. @hgaiser, feel free to review too!)

Hi, @stylishsam! I don’t know if `model.save` works with custom layers. I’ll look into this. Have you tried using `to_json` and `save_weights`?

I agree. Thankfully, it should be straightforward. I haven’t looked their pooling implementation, but I don’t remember the details in the paper being particularly tricky to be implement. Would you...

Hi, @MohmadAyman! We’d love your help. - We still need to write the Mask RCNN model. It should be straightforward and is likely the easiest item to contribute if you’re...

Hi, @MohmadAyman! Because we’re moving the anchor generator into the loss function, you don’t need to worry about the anchors. The generator just needs to yield an image, a list...