topic Re: Export Training Data For Deep Learning The Difference In Brightness of Output Files in ArcGIS Pro Questions

Export Training Data For Deep Learning The Difference In Brightness of Output Files

sukruburakcetin — Tue, 01 Feb 2022 21:20:59 GMT

Hello there,
When I try to use the Image Analyst tool Export Traning Data For Deep Learning, the exported tile outputs' come with an arranged level that their brightness is exaggerated.

What I mean is that the output tiles' color balance or luminance differentiation doesn't suit the original raster file. This is not a big issue since I examined the results are not being affected too much after I executed sample tasks for my deep learning needs. However, I would like to know what is the exact agent that is causing shifts in the brightness of the exported tile files.

I tried all configurations including extension modes, metadata format, reference system. All are leading the same output.

I added a sample region that clarifies the issue as you may be interested to get further info.

Re: Export Training Data For Deep Learning The Difference In Brightness of Output Files

DanPatterson — Tue, 01 Feb 2022 23:15:32 GMT

what symbology are you using for the layer? Do they look the same un clipped/stretched assuming you are using rgb and default symbology

Re: Export Training Data For Deep Learning The Difference In Brightness of Output Files

sukruburakcetin — Wed, 02 Feb 2022 09:36:53 GMT

First of all thank you so much dear @DanPatterson .

To clarify the issue as much as detailed, I used 4 bands .tif imaginary and simply extracted the first 3 bands separately, then used composite band tool to reunite them without making any changes in symbology(default settings).

Then, I used Export Training Data for Deep Learning Tool as with the setup you mentioned in your previous message.

In short, the symbology I used has a percent clip stretch type (min 0,500 to max 0,500) with 0,9 gamma.

After your saying, this might be the issue, but when I want to display the 3 band composited image and compare it with the one that has 4 bands(original specimen), the color balance looks the same. 4 bands original imaginary has also same symbology(percent clip stretch type min 0,500 and max 0,500)

layer_symbology visual is added to the attachments to make it more pronounced. Thanks in advance.

Re: Export Training Data For Deep Learning The Difference In Brightness of Output Files

DanPatterson — Wed, 02 Feb 2022 12:48:40 GMT

you should try the comparison without any clip or stretch is what I was suggesting since the range of values in each image will determine the high and low values and over what range to make adjustments