/bin/pretrained segfaults

[dougsouza]

Hi, I am having trouble doing inference on a GoogLenet (same happens for resnet-50)

$./bin/pretrained --init_net res/googlenet_init_net.pb --predict_net res/googlenet_predict_net.pb --size 224

## Caffe2 Loading Pre-Trained Models Tutorial ##
https://caffe2.ai/docs/zoo.html
https://caffe2.ai/docs/tutorial-loading-pre-trained-models.html
https://caffe2.ai/docs/tutorial-image-pre-processing.html

init-net: res/googlenet_init_net.pb
predict-net: res/googlenet_predict_net.pb
file: res/image_file.jpg
size: 224

image size: [1280 x 751]
scaled size: [381 x 224]
cropped size: [224 x 224]
value range: (-128, 127)
*** Aborted at 1523737624 (unix time) try "date -d @1523737624" if you are using GNU date ***
PC: @    0x1d4f4 caffe2::TypeMeta::Match<>()
*** SIGSEGV (@0x0) received by PID 840 (TID 0x6fcacbd0) from PID 0; stack trace: ***
    @ 0x75b926c0 (unknown)
    @    0x1d4f4 caffe2::TypeMeta::Match<>()
    @    0x1b114 caffe2::Blob::IsType<>()
    @    0x194b4 caffe2::Blob::Get<>()
    @    0x15328 caffe2::run()
    @    0x15ae0 main
    @ 0x75b7c678 __libc_start_main
Segmentation fault

Any ideias if the problem is in the model or in my build of caffe2?

Thanks!

[leovandriel]

Thanks for reporting! Fixed in commit 15530ec3.