search

xuebinqin / DIS

This is the repo for our new project Highly Accurate Dichotomous Image Segmentation
Apache License 2.0
2.11k stars 244 forks source link

Multiple sets of data sets, the accuracy of the last set cannot be improved #112

Open babyta opened 3 months ago

babyta commented 3 months ago

hello author，I used 9 sets of data, using the incremental learning method, starting from one set and slowly increasing it. Later, when I added DIS data, the reorganized data verification accuracy was lower, as shown below _maxF1_0.9685_0.9713_0.9704_0.9713_0.9703_0.9729_0.9709_0.8183_0.9881 0.8183 is the accuracy of the DIS data set. When I use this data alone for training, the accuracy is very high. I suspect it is a problem with the distribution of the rest of my self-built data and the current data.

Mlegbder commented 2 weeks ago

I'm having this problem too, did you find the problem?