Make pakage dnn-proxy-caffe for integration with crowdsource-video-experiments-on-desktop

dividiti / ck-caffe

Collective Knowledge workflow for Caffe to automate installation across diverse platforms and to collaboratively evaluate and optimize Caffe-based workloads across diverse hardware, software and data sets (compilers, libraries, tools, models, inputs):

http://cKnowledge.org

BSD 3-Clause "New" or "Revised" License

193 stars 40 forks source link

Make pakage dnn-proxy-caffe for integration with crowdsource-video-experiments-on-desktop #52

Closed Chunosov closed 7 years ago

Chunosov commented 7 years ago

Extract all lib-dependent code into separate package, which will be compiled into so-file (or dll) and will have well-known set of exported function - preparation model, parallel recognition etc. The app could find this proxy lib via CK, load and do real work through it. Such proxy-lib should be depend only on required things and hide any lib-specific stuff from the desktop app.

psyhtest commented 7 years ago

Please note that similar proxy libs will be required for CK-TensorFlow, CK-TensorRT, etc.

gfursin commented 7 years ago

We decided not to use this method for now but use caffe-classification binaries directly. However, we are not removing this method completely but archived it because it can be very useful for realistic applications. But for trying different libs (caffe-cpu,caffe-opencl,caffe-cuda), it requires lots of API gluing ... But thanks a lot for the first version - it was very useful! I close this ticket!