Question with custom dataset training #11356
Unanswered
kronee0516
asked this question in
Q&A
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
I try to detecting some kinds of animals with custom training
but the result is not that good
hope you guys can point out my mistakes
first i want to say what the problem is
there are 2 things that happen quite common
my dataset have 6 classes and around 16000 images in total (most of them only have 1-2 instances)
and I enlarged the dataset to 80000 with the help of albumentation
but the number of classes and images are not balanced as some kind of animal's image is hard to find
is this a huge issue?
and should i use pretrained model to train my custom model
i rose this question because i would do detection on some video to see if my model valid
some of the video would have human face on it
but the model always detects them as one of my custom class
i don't know if it is related to the pretrained model
or is the reason of unbalanced dataset
for my dataset stated above , 80000 images with 6 classes
what batch size is good for training?
i have tried 32 and 8 ,seems 32 is better but still not good enough
and the image quality.
i am wondering, should i cut out all the target object only?
will this action improve the training result?
last
what else can i tweak for better training result?
Beta Was this translation helpful? Give feedback.
All reactions