FCN Binary Segmentation with Scattered Segments, Is It Overfiting?

[Dean-TianZhang]

Please use the caffe-users list for usage, installation, or modeling questions, or other requests for help.
Do not post such requests to Issues. Doing so interferes with the development of Caffe.

Please read the guidelines for contributing before submitting this issue.

Issue summary

Recently, I am working on a project related with 2-class segmentation(0 for background, 1 for object). And I am using the pretrained caffe model to fine tune my model. But after training my model for 100000 iterations, the prediction performance is really bad with Scattered Segments. I have modified the following part of FCN-8s model:

1. number_output: 21 -> 2
2. add weight filter and bias_filter to each convnet. and choosing "type: gaussian" for weight_filler, "type: "constant" for bias_filler.

And here is a output:

Actually, the grey area in the above image should be continuous.

Does anyone have the same problem? Is it causing by overfitting, since I just use unseen images to predict?

Steps to reproduce

If you are having difficulty building Caffe or training a model, please ask the caffe-users mailing list. If you are reporting a build error that seems to be due to a bug in Caffe, please attach your build configuration (either Makefile.config or CMakeCache.txt) and the output of the make (or cmake) command.

Your system configuration

Operating system:
Compiler:
CUDA version (if applicable):
CUDNN version (if applicable):
BLAS:
Python or MATLAB version (for pycaffe and matcaffe respectively):