search

matterport / Mask_RCNN

Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
Other
24.54k stars 11.68k forks source link

How to train and test at a new class? #1779

Open shanruiyu opened 4 years ago

shanruiyu commented 4 years ago

Hi, thank you for reading my issue!

I have a medical dataset consisting of 1.4k thyroid ultrasound images, there is only 1 class — thyroid nodule. I have converted the dataset into coco format by using pycococreator, and set the number of class to 1+1( including background ) in config file and sample/coco/coco.py. But after training and evaluating, the Average Precisions and Average Recall are 0.000.

I write this issue to ask: How to modify the code to make the model suitable for detecting a new category instead of the 81 classes of coco.

STASYA00 commented 4 years ago

Could you, please, write the parameters you have passed to the configuration and to the model (in the training phase)?