Im trying to build batches for accuracy testing of googlenet model
I have created the lmdb file and corresponding binaryproto file.
I use the ~/caffe/models/bvlc_googlenet/train_val.prototxt file, update the absolute path to lmdb file in the train_val-prototxt file.
I then use below command to run the caffe test command to generate batches.
50 per batch x 1000 iterations = 50000 images
caffe/build/tools/caffe test -gpu 0 -iterations 1000 -model /home/ubuntu/caffe/models/bvlc_googlenet/train_val.prototxt -weights /home/ubuntu/TensorRT-184.108.40.206/data/googlenet/googlenet.caffemodel
I use the default googlenet.caffemodel file in the TensorRT install path as weights (I m assuming its trained for googlenet) .
Below is the snapshot of error i get.
I suspect it is to with resizing/cropping of the images . Any help is highly appreticated.
F0529 11:55:49.607640 4262 net.cpp:774] Cannot copy param 0 weights from layer ‘loss1/classifier’; shape mismatch. Source param shape is 20 1024 (20480); target param shape is 1000 1024 (1024000). To learn this layer’s parameters from scratch rather than copying from a saved net, rename the layer.
*** Check failure stack trace: ***
@ 0x7f330c9da5cd google::LogMessage::Fail()
@ 0x7f330c9dc433 google::LogMessage::SendToLog()