xyuan13 / msrnet Goto Github PK

This MSRA-B dataset is not available on the Internet now, because the url link of this dataset from the author Jian Sun's homepage (jiansun.org) point to 404.

You had created an amazing and large model based on pre-trained VGG model, but you set the same learning rate in the pre-trained models and almost all the following layers in your training. However, the scale of saliency datasets are much smaller than ImageNet dataset, and the results may suffered from overfitting problem. So, how did you solved this problem in you training ?

Hi ,
I had an issue when downloading the model ( Prototxt was ok though). I tried manually and apparently Baidu requires a login certification since the model is quite large. ( Signing up possible only for Chinese phone number holders). Is it possible to share the files over Drive or dropbox to avoid the issue. thanks.

param.modelFile = fullfile(models_root, sprintf('MSRNet-%s.prototxt',model_id));
this line should be rewrite as:
param.modelFile = fullfile(models_root, sprintf('MSRNet-%s.caffemodel',model_id));
Please confirm this.
This mistake would raise error in the line 45.

Hi there,
I was actually interested in the results of the instance saliency segmentation and I think the given Model outputs just the Smap (which is quite good btw). So I was wondering the estimation of the salient contours and the objects proposals is done separately by another model or u have a model combining everything together.

I have read this paper twice, but I can not find any cues about the cost function. Could you give me some hint? My email: shizhoudong at 126.com. Thank you very much.

Hi, I would love to test MSRNet on my dataset. Can you please provide details on how to train MSRNet with a new dataset?