What is considered samples in Tensorflow object detection API?

General Discussion
,
#1

In image classification, it is obvious that images are considered as samples but what is considered as samples in Tensorflow object detection API? Are the samples images or the individual objects in the images?

if an image contains 10 objects, should I consider 10 objects as training samples or just a single image?

I need to know this to effectively set batch size and steps for training.

1 Like
#2

The question is very interesting.
IMO it should be images here as well since what we define in API config as batch_size is number of images while training object detection model.

2 Likes