search

szad670401 / Fast-MTCNN

a casual work about retraining to optimize mtcnn Pnet and ONet. it can achieve 100+fps on CPU with minSize 60 (1920x1080) on intel i7 6700k
203 stars 63 forks source link

about the det2_half.caffemodel #10

Open zyg11 opened 5 years ago

zyg11 commented 5 years ago

Hello, I test these models , and have found the size of det2_half.caffemodel is just 198k,more smaller than the size of origin model which is 398k.But I find the prototxt of det2 is same as original det2's prototxt .So could you tell me how to compress the det2_half.caffemodel to 198k size det2_half.caffemodel?Thank you!

szad670401 commented 5 years ago

the det2_half is same version as the origin mtcnn version.I convert it to fp16 with opencv.

tanghy2016 commented 5 years ago 
>>> import caffe
>>> deploy = "./det2.prototxt"
>>> model = "./det2_half.caffemodel"
>>> net = caffe.Net(deploy, model, caffe.TEST)

WARNING: Logging before InitGoogleLogging() is written to STDERR W0124 13:46:58.335070 14075 _caffe.cpp:122] DEPRECATION WARNING - deprecated use of Python interface W0124 13:46:58.335242 14075 _caffe.cpp:123] Use this instead (with the named "weights" parameter): ... I0124 13:47:03.393421 14075 net.cpp:283] Network initialization done. I0124 13:47:03.394093 14075 net.cpp:761] Ignoring source layer data_pos I0124 13:47:03.394105 14075 net.cpp:761] Ignoring source layer data_part I0124 13:47:03.394112 14075 net.cpp:761] Ignoring source layer data_neg I0124 13:47:03.394119 14075 net.cpp:761] Ignoring source layer pnet_data I0124 13:47:03.394124 14075 net.cpp:761] Ignoring source layer pnet_label I0124 13:47:03.394129 14075 net.cpp:761] Ignoring source layer slicer_data I0124 13:47:03.394134 14075 net.cpp:761] Ignoring source layer nulllabel F0124 13:47:03.394157 14075 blob.cpp:507] Check failed: count == proto.data_size() (756 vs. 0) Check failure stack trace: Aborted (core dumped)

ALONGNEU commented 5 years ago 
>>> import caffe
>>> deploy = "./det2.prototxt"
>>> model = "./det2_half.caffemodel"
>>> net = caffe.Net(deploy, model, caffe.TEST)

WARNING: Logging before InitGoogleLogging() is written to STDERR W0124 13:46:58.335070 14075 _caffe.cpp:122] DEPRECATION WARNING - deprecated use of Python interface W0124 13:46:58.335242 14075 _caffe.cpp:123] Use this instead (with the named "weights" parameter): ... I0124 13:47:03.393421 14075 net.cpp:283] Network initialization done. I0124 13:47:03.394093 14075 net.cpp:761] Ignoring source layer data_pos I0124 13:47:03.394105 14075 net.cpp:761] Ignoring source layer data_part I0124 13:47:03.394112 14075 net.cpp:761] Ignoring source layer data_neg I0124 13:47:03.394119 14075 net.cpp:761] Ignoring source layer pnet_data I0124 13:47:03.394124 14075 net.cpp:761] Ignoring source layer pnet_label I0124 13:47:03.394129 14075 net.cpp:761] Ignoring source layer slicer_data I0124 13:47:03.394134 14075 net.cpp:761] Ignoring source layer nulllabel F0124 13:47:03.394157 14075 blob.cpp:507] Check failed: count == proto.data_size() (756 vs. 0) Check failure stack trace: Aborted (core dumped)

Have you solve this problem yet? I find this too. @tanghy2016