ignacio-rocco/detectorch

Get right result of bbox using eval_mask_FPN code but wrong result of segmentation?

Opened this issue · 5 comments

JackHenry1992 commented

Thanks for sharing your greate work.
I have tested your mask_rcnn_FPN code, and got the right result of bbox detecting, but wrong result of seg.
I have pasted the result of eval_mask_FPN AP as follows:

  1. bbox
    image
  2. seg

image

And the demo_mask_FPN also shows a wrong result as follows:

image

  1. Furthermore, I test the mask_rcnn_c4_2x-resnet50 model, and the result is same as yours.
    image

So is there something wrong with mask_rcnn_FPN ?

ignacio-rocco commented

Hi, which pretrained model are you using?
I'm using model 35861858 and it's working fine for me.

JackHenry1992 commented

Hi, @ignacio-rocco, I have tested these two models
image
and can't get the right result.
I just test the 35861858 model again, and also get the wrong semantic-mask result but right bbox.

cherryxiongyw commented

@JackHenry1992 Hi，do you solve the problem？When I run domo_FPN.ipynb，I got the same wrong result。

ignacio-rocco commented

Which python & pytorch version are you using?

cherryxiongyw commented

@ignacio-rocco python=3.5 pytorch = 0.3.1