- Caffe
- CUDA and CuDNN suitable to Caffe
- opencv
Please follow the parameter instructions in solver to set the proper learning rate. In the solver, we provide with the form as follows, dataset task [task ] lr. And the condition that test_iter x test_batch_size=test_data_size should be satisfied.
After set the right parameters, please doenload the pretrained alexnet model in ImageNet here to models/bvlc_reference_caffenet. The command for running code is
./build/tools/caffe -solver models/cdan/solver_cdan_alex.prototxt -weighths models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel -gpu gpu_id
- We only implement the alexnet version for caffe due to memory consumption of caffe resnet.