I was able to convert a mobilenet_v2 model’s tflite file to kmodel and the size of the kmodel file is 21MB.
tflite to kmodel conversion command:
./ncc/ncc compile tflite_graph.tflite tflite_graph.kmodel -i tflite -o kmodel --dataset tenten/ --inference-type float
I ran an inference command from nncase compiler :
./ncc/ncc infer tflite_graph.kmodel op/ --dataset tenten/
and i am getting a binary file for every image given in the input dataset.
Is there a way to know the prediction score and label for each image?
Could someone suggest a way to reduce the size of this model to less than 16MB as the flash memory of sipeed mcu is 16MB?
Thanks in advance!