search

eric612 / MobileNet-YOLO

A caffe implementation of MobileNet-YOLO detection network
Other
865 stars 442 forks source link

Cityscape inference issue #237

Open didibaba opened 4 years ago

didibaba commented 4 years ago

Dear author,

I was using the default settings and pretrained model with c++ given here https://github.com/eric612/MobileNet-YOLO/tree/master/data/cityscapes. The result seems to be wrong. However, I can obtain correct result with detect_and_segmentation.py. Is there anything that I missed to set in the program?

Thanks a lot!

Screenshot from 2020-02-29 00-26-48 Screenshot from 2020-02-29 00-27-10

eric612 commented 4 years ago

It looks like a resolution problem , the default mean , scale and input resolution were different between bdd100k and cityscapes model . The default parameter is 'bdd100k' in detect_and_segment.py

didibaba commented 4 years ago

Just setting the seg_resize_scales to 8 solved my problem.

didibaba commented 4 years ago

@eric612 I tried to change the seg_resize_scales to 4, but error occured.

65 yolo_seg.cpp:33] Check failed: bottom[0]->count() == bottom[1]->count() (46208 vs. 184832) YoloSeg layer inputs must have the same count.

Could you show me how to make this work? Thanks.