Exported tensorflow lite is as large as tensorflow model

@@DemetrioScarfone-7181 There are several techniques you can use to reduce the size of your TensorFlow Lite model for offline image classification:

Quantization: You can use post-training quantization to reduce the precision of the model weights from 32-bit floating point to 8-bit integers. This can significantly reduce the size of the model without sacrificing too much accuracy. You can also use quantization-aware training to train the model with quantization in mind, which can further reduce the size of the model.

Pruning: You can use weight pruning to remove the least important weights from the model. This can reduce the size of the model without sacrificing too much accuracy. You can also use structured pruning to remove entire filters or channels from the model, which can further reduce the size of the model.

Knowledge distillation: You can use knowledge distillation to train a smaller student model to mimic the behavior of a larger teacher model. This can reduce the size of the model while maintaining most of its accuracy.

Architecture optimization: You can use architecture optimization techniques such as neural architecture search to find a smaller and more efficient model architecture that achieves similar accuracy to your current model.

Compression: You can use compression techniques such as weight sharing, Huffman coding, or arithmetic coding to further reduce the size of the model.

You can try one or more of these techniques to reduce the size of your TensorFlow Lite model to below 10MB. Keep in mind that reducing the size of the model may also affect its accuracy, so you may need to experiment with different techniques and find the right balance between size and accuracy for your use case.