MaKaNu
commented
2 years ago I am working at the moment at a different dataset, which needed a lot of tweaks to get them training with the full implementation. At the moment I am able to calculate the UCM for my dataset and I am halfway through the pareto_choose_point script.
To follow my steps analyze the following files:
root_dir: Add your own dataset or modify one of them.
ground_truth: I added my dataset, since my gt is based on pascal
get_image: I needed to add a check for my dataset, because it was a BW-image and uint16
params: Update line 18 to 30 corresponding to your dataset.
(optional) rgb2pascal_voc: This function transfers a rgb image to pascal voc indexed png. There is also a script which demonstrate how it works. BE CAREFULL THE STANDARD USE OF THE FUNCTION OVERWRITES YOUR IMAGES
Additional there was a recent PR which is merged into master, which allows the usage of modern MATLAB function above MATLAB 2015A.
Status of my analysis:
I personally struggle at the moment, where I set the number of classes represented in my dataset to finish pareto learning.
Dear author： Thank you very much for the open source code, I still have some questions to ask. Can I re-train MCG on a new dataset (e.g., natural scene text images) to get proposals or segmentation results? Looking forward to your reply!