I am using a SSD object detection model to run on android using jit trace, developed in PyTorch 1.3.
2 days back, I was using the nightly pytorch_android:1.5.0-SNAPSHOT it was working fine and I was getting the expected output. But today the app is crashing and giving the following error Fatal signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0x0 in tid 17354.
After this I tried other versions also.
pytorch_android:1.4.0 and pytorch_android:1.4.0-SNAPSHOT : This is giving different output. For the same input Tensor, output tensors in python and android are coming out to be different.
pytorch_android:1.3.0 : com.facebook.jni.CppException: false CHECK FAILED at ../c10/core/Backend.h (tensorTypeIdToBackend at ../c10/core/Backend.h:106)
Very confused with what’s going on! Can I revert back to the nightly build which was 2 days old?