2

I am trying to implement object detection on satellite images. I have a annotated dataset, but the images are large and the model accepts only 416 x 416 size inputs. How can I pass small parts of the image in the network, ensuring that the annotations are retained. Also, how to merge these results at test time?

vvv
  • 461
  • 1
  • 5
  • 14
  • do you want to cut the image into small chunks and feed all chunks into the network or do you want to cut out one part and feed this into the network? – Jonathan R Jun 01 '18 at 09:48
  • Please provide some extra information regarding the actual size of satellite images, their type (if preprocessing is required to get to common format), the kind of annotation you have, the model you are going to use, the number of classes etc – Eypros Jun 01 '18 at 09:48
  • @JonathanR the former – vvv Jun 01 '18 at 10:24

1 Answers1

2

crop them with a small padding to prevent border effects and merge them back. you can see how this is done over here, line 552.

https://github.com/lopuhin/kaggle-dstl/blob/master/train.py

Zaw Lin
  • 5,629
  • 1
  • 23
  • 41