search

soeaver / caffe-model

Caffe models (including classification, detection and segmentation) and deploy files for famouse networks
MIT License
1.28k stars 626 forks source link

Difference between inception-v4-3x3 model and inception-v4 model about RFCN #46

Closed GT9505 closed 6 years ago

GT9505 commented 6 years ago

Hi,soeaver. I notice when you use RFCN train inception v4 model, you trained two kind of model: inception v4 and inception v4 3x3. Could you tell me the difference between inception v4 and inception v4 3x3.

soeaver commented 6 years ago

I replace the 1x3 and 3x1 conv with 3x3 conv in inception-v4 for dilated conv, that's the inception-v4-3x3

GT9505 commented 6 years ago

When I try to use RFCN train inception v4 with VOC0712 trainval, I only get 77.33mAP on VOC2007 test. I notice the solver.prototxt and rfcn_end2end_ohem.yml you provided is used to train ResNet-101 v2. Is there difference about the two files when training the two kind of models

soeaver commented 6 years ago

There is no difference between the two files.

GT9505 commented 6 years ago

Thanks a lot!