Distractor-aware Siamese Networks for Visual Object Tracking Inference in C++ powered by Caffe2
For more detail information:Please visit the original Pytorch repo:https://github.com/foolwood/DaSiamRPN
python2.7 (for convert model only)
pytorch == 0.3.1 (for convert model only)
numpy (for convert model only)
caffe2
Eigen
opencv
CUDA
if you just use my converted model, python2.7 pytorch and numpy is not Necessary。
The Model is Convert from the original Pytorch not by the ONNX but by a python script(in script dir). Because the DaSiamRPN's inference is a Dynamic graph Not Static graph.As description in onnx it has limit to export Dynamic graph.
Converted model for SiamRPN's Download link: 链接: https://pan.baidu.com/s/1GfbmHmR0HjjAK0yU5JhGpA 提取码: zbev
Other note: This Repo uses the older Caffe2,the Newest Caffe2 may be not work well(because it has many changes in recent mouths, For Example the newest caffe2's Tensor class has been removed the template etc.). One of the Old Caffe2 can be clone from https://github.com/zj463261929/caffe2_old