Epoch vs Steps_per_epoch vs Batch_size

[Densus100]

Hello guys, I'm trying to re-train the Mask RCNN to fit into my dataset, but i stuck at the config and really confused about some parameters. Please anyone that maybe able to help me distinguish the difference between :

1. EPOCH
2. STEPS_PER_EPOCH
3. BATCH_SIZE

Explanation with example like the numbers of the dataset and how many iteration that the models do between those parameters will be much appreciated. Thank you :)

[raulperezalejo]

Hello, this is my understanding:

EPOCH and STEPS_PER_EPOCH: can be found here as an input parameter of the fit method.

BATCH_SIZE: This is calculates this way: BATCH_SIZE =IMAGES_PER_GPU * GPU_COUNT GPU_COUNT is simply the amount of GPU you have, for example is colab is only 1 IMAGES_PER_GPU: is the amount of images the CPU is going to process each time. In my case ill send my model one picture each time, in this case i set this number to 1

hope it helps

[Densus100]

Thank you @raulperezalejo