Feature Request: Add Patch-Based Inference Support (Inspired by MCUNetV2)

[Josiah-MCS]

Feature Request: Add Patch-Based Inference Support (Inspired by MCUNetV2)

Problem Statement

TensorFlow Lite Micro (TFLM) currently lacks support for patch-based inference, as introduced in MCUNetV2. This technique processes input images in smaller patches sequentially, reducing peak memory usage, thus enabling inference on higher resolution images on resource-constrained devices like microcontrollers.

References

• MCUNetV2: Memory-Efficient Patch-Based Inference