EXC_BAD_ACCESS error on IOS

[maslovay]

I'm using custom SSD model (I tried my own and three other models that were trained by users)

They work fine in Android, but in IOS it fails here:

if (interpreter->AllocateTensors() != kTfLiteOk) {
    callback(@[@"Failed to allocate tensors!"]);
  }

With EXC_BAD_ACCESS

When I use ssd_mobilenet.tflite from model Zoo IOS does not throw error

[BILLXZY1215]

Same issue here. Any workaround?

[ananta]

Any solution over here? I'm facing the same issue :)

[BILLXZY1215]

Hey guys, are you using the customized tflite model? It works fine with my machine when I use the default model, but not for the customized models. I think when generating tflite files, the version of tensorflow-lite should be specified. This repo only accepts tensorflow-lite r1.12, which means the tflite file generated by other versions might not match the interpreter. In this case, you may need to replace these files:

ios/Pods/TensorflowLite/Frameworks/tensorflow_lite.framework/Headers/Tensorflow to a suitable version (e.g. tensorflow r2.8)
ios/Pods/TensorflowLite/Frameworks/tensorflow_lite.framework/Headers/flatbuffers to a suitable version (e.g. flatbuffer 1.12, which will match tf r2.8)