I used Vision AutoML to train a dedicated model. My dataset contain panoramic pictures (3000x1500px). This model works fine on tensorflow.lite (2.5.0). But I want to load it on Coral Edge TPU.
When I use edge_compiler, fail with this error :
Edge TPU Compiler version 15.0.340273435 loc("reshape_output"): error: 'tfl.reshape' op quantization parameters violate the same scale constraint: !quant.uniform<u8:f32, 0.43928399682044983:111> vs. !quant.uniform<u8:f32, 0.4234168529510498:111>
Internal compiler error. Aborting!
Any ideas ?