Label and export training sample for multiple images

[Anonymous User]

I need to label objects in multiple images and export the training data to ArcGIS Pro. I would like to train the SSD model using objects labelled in different images. What is the correct workflow?

[Tim_McGinnes · Accepted solution]

What type of images are they?

For aerial or satellite imagery the way I do it is to create a mosaic dataset of the images with a single featureclass for the labels. Then use the mosaic dataset as input to the Export Training Samples tool.

If you are using normal real world digital photos, then it would probably be better to use a specialized image labelling program. There are lots to choose from, for example LabelImg or VoTT. If you are doing Object Detection, then tool should output Pascal VOC format. There may be a bit of fiddling around required with formats etc. if you go this way.

[Anonymous User]

Thanks Tims,

i have satellites images

[Anonymous User]

I have another doubt. We have created a mosaic dataset but the images are from different times on the same area.

In this case how can I export training samples from overlapping images?