Mendix ML Kit model mapping generates wrong output types for specific shapes of MaskRCNN
0
Hi all, I'm currently using an ML model for object detection with Mendix ML Kit with dynamic tensors. When I specify the input image to the shape [3,256,256] this works fine and the output is mapped as base64 encoded data. However there seems to be a problem with the output shape. The value calculated there doesn't really align with the actual amount of data returned by the model. The problem I now have comes up when changing the input shape to [3,256,256] because then the output changes not only in shape but also in type. The second and third value are now no longer an array encoded as base64. Is there a way to fix that behaviour within Mendix? Why does the output type change at all? Would be happy to hear any suggestions on that problem. regards, Fabian P.S. actually 10.6.2 but the forum always lacks behind with version numbers