Training Job Failed "ClientError: Mask image is not a 8-bit"

Question

Hello,  
  
I am working on a Semantic Segmentation training job.  I have used the Ground Truth semantic segmentation labeling GUI and have labeled about 900 images.  Everything with the Ground Truth Labeling job seemed to work just fine.  However I noticed the output images in the s3 output path are of type .png.    
  
I am now trying to build a training job using the Pipe - augmented manifest input format.  When I kick off the job, after a few minutes I see this error: **"ClientError: Mask image is not a 8-bit single channel image. Please check the label image dataset"**  
  
I downloaded the output images or "masked images" to photoshop and they do appear to be 8bit "index color".  
  
1) If I used the built in Semantic Segmentation - Ground Truth Labeling job, how would I end up with the incorrect output format?  
2) I don't remember specifying or having the option of output image format.  Did I miss that somewhere?  
  
Here are a couple line items from the output.manifest file

```
{"source-ref":"s3://rigbypotato-2/segmentation-validation/MyImage - 20190201122850904.png","potato-features-validation-ref":"s3://rigbypotato-2/segmentation-validation/output/potato-features-validation/annotations/consolidated-annotation/output/0_2020-04-14T02:38:26.487862.png","potato-features-validation-ref-metadata":{"internal-color-map":{"0":{"class-name":"BACKGROUND","hex-color":"#ffffff","confidence":0.79952},"1":{"class-name":"pressure-bruise","hex-color":"#2ca02c","confidence":0.79952},"2":{"class-name":"potato","hex-color":"#1f77b4","confidence":0.79952},"3":{"class-name":"old-bruise","hex-color":"#ff7f0e","confidence":0.79952},"4":{"class-name":"black-skin","hex-color":"#d62728","confidence":0.79952},"5":{"class-name":"green","hex-color":"#9467bd","confidence":0.79952},"6":{"class-name":"scab","hex-color":"#8c564b","confidence":0.79952}},"type":"groundtruth/semantic-segmentation","human-annotated":"yes","creation-date":"2020-04-14T02:38:26.563488","job-name":"labeling-job/potato-features-validation"}}
{"source-ref":"s3://rigbypotato-2/segmentation-validation/MyImage - 20190201122852921.png","potato-features-validation-ref":"s3://rigbypotato-2/segmentation-validation/output/potato-features-validation/annotations/consolidated-annotation/output/1_2020-04-14T02:21:27.433381.png","potato-features-validation-ref-metadata":{"internal-color-map":{"0":{"class-name":"BACKGROUND","hex-color":"#ffffff","confidence":0.82409},"1":{"class-name":"pressure-bruise","hex-color":"#2ca02c","confidence":0.82409},"2":{"class-name":"potato","hex-color":"#1f77b4","confidence":0.82409},"3":{"class-name":"old-bruise","hex-color":"#ff7f0e","confidence":0.82409},"4":{"class-name":"black-skin","hex-color":"#d62728","confidence":0.82409},"5":{"class-name":"green","hex-color":"#9467bd","confidence":0.82409},"6":{"class-name":"scab","hex-color":"#8c564b","confidence":0.82409}},"type":"groundtruth/semantic-segmentation","human-annotated":"yes","creation-date":"2020-04-14T02:21:30.556262","job-name":"labeling-job/potato-features-validation"}}
```

Thanks  
  
Edited by: tetontech on Apr 16, 2020 7:07 PM

Answer

Total user error.  
  
My problem was that I was only giving the semantic segmentation algorithm 1 attribute from my output.manifest file which was "source-ref". This algorithm of course needs 2 attributes from the the output.manifest file which are the source images **and** masked images.  Once I provided the second attribute "potato-features-validation-ref", everything worked fine.    
  
From the few lines of the output.manifest file I posted earlier, you will notice these two attributes.  
  
hope this helps someone else.

Training Job Failed "ClientError: Mask image is not a 8-bit"

Relevanter Inhalt