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

Question about result of toy data. #103

Open MengzhangLI opened 4 years ago

MengzhangLI commented 4 years ago

Hi, I'm a beginner of MedicalDetectionToolkit, this is a very amazing framework for medical object detection. Thanks for your contributions to community!

When using MDT on 2D toy data by implementing: python experiments/toy_exp/generate_toys.py numerical results is a little bit wrong: the trend of loss is not stable: monitor_0_0

The visulization of 2D toydata shows it has many bbox candidates.(NMS not work?) I didn't change any parameters on config.py, do I forget or misuse some parameters of MDT? WeChat Screenshot_20200515045928

Best,

Mengzhang LI

Satantago commented 1 year ago

Hello! I'm having the same questions, did you manage to find convincing answers?

MengzhangLI commented 1 year ago

Hello! I'm having the same questions, did you manage to find convincing answers?

It is mainly because of small object with/without hole in the center is difficult to discriminate. If I remember well, if you make the inner hole bigger, the validation results would become better. You can have a try.

Best,