search

pjreddie / darknet

Convolutional Neural Networks
http://pjreddie.com/darknet/
Other
25.75k stars 21.33k forks source link

Instance_Segmenter? #1064

Open sayanmutd opened 6 years ago

sayanmutd commented 6 years ago

How do i use instance_segmenter? Could you specify the cfg and weight files and what is the expected output? @pjreddie @lilohuang

aimhabo commented 6 years ago

The same problem, I don't know how to use the iseg layer

sayanmutd commented 6 years ago

I don't know which configuration and weight files to use for running the instance segmenter demo? @aimhabo @pjreddie @AlexeyAB

aimhabo commented 6 years ago

@sayanmutd As far as the files are currently seen, it is not a complete applicable branch. but you can try to fellow the parser.c and data.c to make any file

sayanmutd commented 6 years ago

@aimhabo Yeah, You are right... that can be done but as it was given in the present darknet i thought some examples might be there... and without proper example you cannot guarantee about what would be expected outcome.

aimhabo commented 6 years ago

2018-08-28 13-18-06 2018-08-28 13-16-30 2018-08-28 13-16-08 @pjreddie hello, I try it directly. It seems that this modification method is not possible. It memory wrong with iseg_layer in forward_network_gpu. How could I fix it?

ArtyZe commented 6 years ago

I have changed the darknet use to image segmentaion, if you have interest, you can see my code: https://github.com/ArtyZe/yolo_segmentation. If you think it's useful, please give me a star :)

aimhabo commented 6 years ago

I can run iseg_layer now, but loss starts from 550,000, and blow-up to NaN after 500 times (64batch/32sub, and it cost 120s+ on GTX1070 every time...). Does it ordinary? what's the meaning of 'ids'? Why set it to 32 in default?

aimhabo commented 6 years ago

now problem becomes "l.delta does not change effectively, causing loss to not converge (although not divergent)"

zouwen198317 commented 6 years ago

I have changed the darknet use to image segmentaion, if you have interest, you can see my code: https://github.com/ArtyZe/yolo_segmentation. If you think it's useful, please give me a star :)

Hi, where r u cfg ? and could you share your cfg file?

abhigoku10 commented 5 years ago

@zouwen198317 did u get the cfg and weight file @ArtyZe can you share the cfg and weight file

ArtyZe commented 5 years ago

You can give me your personal emai ：）

At 2019-06-21 14:45:16, "abhigoku10" notifications@github.com wrote:

@zouwen198317 did u get the cfg and weight file @ArtyZe can you share the cfg and weight file

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

abhigoku10 commented 5 years ago

@ArtyZe abhigoku10@gmail.com is the mail id

sicarioakki commented 5 years ago

@sayanmutd @aimhabo How have you guys used instance_segmenter.c with YOLO? Basically, how is the instance_segmenter.c called with YOLO?

abhigoku10 commented 5 years ago

@ArtyZe thanks for sharing the model !!!!!!! whats the command to run the instance segmenter model

sicarioakki commented 5 years ago

@pjreddie Could you please elaborate as to how the module 'instance_segmenter.c' can be used with YOLO?