-1

Wrong number of CSV columns for AUTOML vision and no documentation on the correct format

I'm trying to use AutoML to train some images I've annotated. It complains that I have the wrong number of columns in the CSV I wrote (saying it should be 11 rather than 9). However, all the examples I see of the CSV are 9 columns. I feel like the format has changed and they didn't update the documentation. Thanks for your help in advance.

Details:

This is their example from their document:

[set,]image_path[,label,x1,y1,,,x2,y2]
TRAIN,gs://My_Bucket/sample1.jpg,cat,0.125,0.25,,,0.375,0.5
VALIDATE,gs://My_Bucket/sample1.jpg,cat,0.4,0.3,,,0.55,0.55
TEST,gs://My_Bucket/sample1.jpg,dog,0.5,0.675,,,0.75,0.875

Here is a line from my test data:

TRAIN,gs://mytensorflowdata/CanVideo 50.jpg,sodacan,0.378125,0.10138888888888889,,,0.61796875,0.8708333333333333

I also tried without the bounding boxes like:

TRAIN,gs://mytensorflowdata/CanVideo 50.jpg,sodacan

Here is the error message AutoML gives me:

Error: gs://mytensorflowdata/labels.csv line 13: Expected 11 columns, but found 3 columns only.
user1335325
  • 67
  • 6

1 Answers1

1

I just ran into the same problem. You are right that they haven't updated the documentation everywhere. This page does show 11 columns though: https://cloud.google.com/vision/automl/object-detection/docs/csv-format. It seems like they added the option to provide all 4 corners of the bounding box. The new example is:

TRAIN,gs://folder/image1.png,car,0.1,0.1,,,0.3,0.3,,
TRAIN,gs://folder/image1.png,bike,.7,.6,,,.8,.9,,
UNASSIGNED,gs://folder/im2.png,car,0.1,0.1,0.2,0.1,0.2,0.3,0.1,0.3
TEST,gs://folder/im3.png,,,,,,,,,

So you need to add two empty columns at the end of your document, like so:

TRAIN,gs://mytensorflowdata/CanVideo 50.jpg,sodacan,0.378125,0.10138888888888889,,,0.61796875,0.8708333333333333,,
Sascha
  • 41
  • 3