Search code examples
pythontensorflowdetectionyolo

Is there a limit about image size when we train custom object with already trained models?

I already trained ssd_mobilenet_v2_coco with my custom data set on tensorflow. Also I trained YOLO with my data set too. I solved all problems and they work.

I encounter a problem with both models. When my data set includes images with more size than 400kb, the trained models do not work. Some times "allocation of memory" problem occurs. I solved them with changing parameters(batch size etc.). But I still don't know whether there is a limit of image size when we are preparing data set?

Why more than 400kb images are problem for my system? My question is not about pixel size,it's about image file size.

Thanks...

System Info

  • Nvidia RTX 2060 6G
  • AMD Ryzen 7
  • 16GB DDR4 2600Mhz Ram
  • Cuda:10.0
  • CudNN:7.4.2(I also try diffrent versions and same results occur)
  • Tensorflow:2
Solution

  • Providing the solution here (Answer Section), even though it is present in the comment section (Thanks dasmehdix for the update) for the benefit of the community.

    No, there is no a limit of image size that we use to train our model.