Open engahmed1190 opened 6 years ago
@engahmed1190 i am having the same problem. were you able to solve this?
hi @argman pre-trained file almost always leaves the single character from the images. can you please help in providing some insight on how to re-train the model to address this problem.
@sg77492 , still struggling I have tried another dataset but results are bad
what do you think @zxytim @argman
Try merge quadrangles using "maximum" of scores instead of average.
On Tue, Jan 2, 2018 at 12:26 AM engahmed1190 notifications@github.com wrote:
@sg77492 https://github.com/sg77492 , still struggling I have tried another dataset but results are bad
what do you think @zxytim https://github.com/zxytim @argman https://github.com/argman
— You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub https://github.com/argman/EAST/issues/77#issuecomment-354662102, or mute the thread https://github.com/notifications/unsubscribe-auth/AA2QT6_k5l9hI0MgTRc_UBhhEa676_PRks5tGQdOgaJpZM4QyCY- .
Hi @zxytim Do you mean the merging done in the locality_aware_nms
def weighted_merge(g, p):
g[:8] = (g[8] * g[:8] + p[8] * p[:8])/(g[8] + p[8])
g[8] = (g[8] + p[8])
return g
or something else happens in icdar file
i am having data looks like that the result from pre-trained model is not bad but i would like to collect more from the image
result
what will be the attributes of the training in this case , how i can get better results