search

ijkguo / mx-rcnn

Parallel Faster R-CNN implementation with MXNet.
Other
669 stars 292 forks source link

Issues with pretrained models #71

Closed AleBasso80 closed 6 years ago

AleBasso80 commented 7 years ago

Hello, I'm running the script demo.py. First I tried to use the pretrained models, but I got the error "rpn_conv_3x3_weight not initialized". Then I tried to use "get_vgg_rcnn" function instead of the default "get_vgg_test" and I got the error "InferShapeKeyword argument name im_info not found". Same issues with the resnet and with the converted caffe py-faster-rcnn model. Instead, if I use the final-0000.params model, everything is fine. Can someone help me to figure it out?

Thanks!

ijkguo commented 7 years ago

faster rcnn is a finetune process from vgg imagenet to voc. need to use the finetuned model final-0000.params model