This repository was a fork of BVLC/caffe and includes the upsample, bn, dense_image_data and softmax_with_loss (with class weighting) layers of caffe-segnet (https://github.com/alexgkendall/caffe-segnet) to run SegNet with cuDNN version 5.
I want to use your model to train my task. And I just replace your dataset with mine (I also change the output to 2). When I train the model, I found if I loaded the pretrianed model (e.g., vgg16), there would be a constant loss (5.723e6). And the results rea always the same class.
Hi,
I have a problem about training.
I want to use your model to train my task. And I just replace your dataset with mine (I also change the output to 2). When I train the model, I found if I loaded the pretrianed model (e.g., vgg16), there would be a constant loss (5.723e6). And the results rea always the same class.
Do you have any idea about the problem?
Thanks,