search

MIC-DKFZ / medicaldetectiontoolkit

The Medical Detection Toolkit contains 2D + 3D implementations of prevalent object detectors such as Mask R-CNN, Retina Net, Retina U-Net, as well as a training and inference framework focused on dealing with medical images.
Apache License 2.0
1.31k stars 297 forks source link

pred_example_{}.png what do the additional colours mean? #77

Closed sophie-isobel closed 4 years ago

sophie-isobel commented 5 years ago

Hi @pfjaeger , thank you for sharing this great repository! My predexample{}.png (for Mask R-CNN) is returning additional coloured boxes that are purple and white, see below:

image

I cannot find the purple and white boxes in the key in the read.me file, may i ask what they mean or why they may be appearing? image

I am using the brats dataset (with 4 modalities shown above), with only one foreground class, and I am training Mask R-CNN with a subset of 20 images until i get the model working correctly. I am training 3 epochs and on the 2nd epoch of the 2nd fold the model is getting stuck, would you have any advice on why this might be happening?

UPDATE: I am now training on a server and it is no longer getting stuck, but there is still an issue with the extra boxes in the pred_example images for Mask R-CNN, see below, notice that it is also not plotting the prediction box in the bottom row. Please let me know what could be causing this, thank you in advance!

image

zoomed in to the issue:

image

thank you, Sophie

CodingHHW commented 5 years ago

in medicaldetectiontoolkit/default_configs.py, line 96-98,

color specifications for all box_types in prediction_plot.
    self.box_color_palette = {'det': 'b', 'gt': 'r', 'neg_class': 'purple',
                              'prop': 'w', 'pos_class': 'g', 'pos_anchor': 'c', 'neg_anchor': 'c'}

means blue------>detection box red------->ground trues purple---->neg_class white----->proposal box from RPN green----->pos_class cyan------>anchor

Gregor1337 commented 4 years ago

Thank you, @CodingHHW, for answering this!