Welcome to OStack Knowledge Sharing Community for programmer and developer-Open, Learning and Share
Welcome To Ask or Share your Answers For Others

Categories

0 votes
511 views
in Technique[技术] by (71.8m points)

regression - Deep learning: howto handle variation in input size

I would like to do a regression Task (find x,y position of an Object in some images with size of approx 150x 70 pixel). Therefore I have about 2000 annotated images (I can annotate more maybe up to 20000). My first thought was to use CNN's but the size of my input data is not consistent and I don't want to lose features by resizing or cropping the images. What do you think? Is there a better way than using CNN's? How do I handle this variation of image sizes?


与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…
Welcome To Ask or Share your Answers For Others

1 Answer

0 votes
by (71.8m points)

ok, here is the thing, if you don't want to rescale or crop. then you won't be able to do the batch processing.

you can scale your input image to one size. just make sure that all your training images are of the same size and similar transformations needs to be done to your ground truth (bound box).

Big problem: I don't wanna apply the transformation to the bounding box and it's complex to do?

Here is a simple library Albumentations that can do all that for you. you need not to worry about the maths- Library - https://github.com/albumentations-team/albumentations


与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…
Welcome to OStack Knowledge Sharing Community for programmer and developer-Open, Learning and Share
Click Here to Ask a Question

2.1m questions

2.1m answers

60 comments

56.7k users

...