Paper "A Plug-and-play Scheme to Adapt Image Saliency Deep Model for Video Data"
The code is tested on windows 10 with MATLAB R2016b, CUDA 8.0
About Our Saliency Maps： Baidu Drive：https://pan.baidu.com/s/1CSixGYStD-i23bOGwxGXYA sd5a
Installation:
Set up our Caffe version, check that Caffe MATLAB wrapper is set up correctly and make sure the path of file "caffe_.mexw64" has been correctly set up.
The experimental model RADF is available in this code and we revise the files based on the original RADF Caffe verison, e.g., "euclidean_loss_layer.cu". Thus, using other Caffe versions will lead to wrong detection.
Usage:
- Our caffemodels can be downloaded from Google Drive. Please download them and put them to the identical path of the file "bat_RADF.m" .
MotionSaliencyDetector: https://drive.google.com/file/d/1oKrQeADuJADcvLaS3GVeL73vqXVV8_HU/view?usp=sharing
RADF original model: https://drive.google.com/file/d/1MGe6CtYoGQYJJYYggAKRjqfN8A50VMt7/view?usp=sharing
-
Video sequences should be copied to the path ".\Sequence\original" with "jpg", "png" or "bmp" file formats.
-
Then, you can use file "bat_RADF.m" to predict saliency maps.
It should be also noted that this code is NOT allowed for commercial purposes.
Thank you
Best Regards