EON Compiler consistently fails on XIAO ESP32-S3 Sense

[sara.aras]

Board: Seeed XIAO ESP32-S3 Sense (OV2640, 8MB PSRAM, 8MB Flash)
Arduino-ESP32 core: v2.0.17
Model: MobileNetV2 96x96, alpha 0.35, quantized int8, image classification (4 classes)

I’m seeing a reproducible crash with the EON Compiler export that I’ve now confirmed on two independent datasets, same architecture, same board.

Error on serial monitor:

ERR: Failed to allocate persistent buffer of size 96, does not fit in tensor arena and reached EI_MAX_OVERFLOW_BUFFER_COUNT
Guru Meditation Error: Core 1 panic’ed (StoreProhibited)

I’ve already tried, in tflite_learn__compiled.cpp:

kTensorArenaSize × 1.5 and beyond
EI_MAX_SCRATCH_BUFFER_COUNT raised from 72 to 128
EI_MAX_OVERFLOW_BUFFER_COUNT raised from 10 to 64

Switching to standard TensorFlow Lite (non-EON) with the same arena-size fix works reliably.

Is this a known incompatibility between the EON Compiler’s static memory planner and ESP32-S3?

I’d like to reference an accurate technical explanation in my thesis rather than just documenting the workaround.
Thank you for any help.

[AIWintermuteAI]

Hi @sara.aras !
Thanks for taking the time to report the issue. It is indeed documented here

Run C++ library on Espressif ESP-EYE (ESP32) - Edge Impulse Documentation

and

ESP32-S3-CAM:ERR: Failed to allocate persistent buffer of size 576, does not fit in tensor arena and reached EI_MAX_OVERFLOW_BUFFER_COUNT

Hi, @2309653 ! TLDR: use non-EON model [image] modify the kTensorArenaSize and /or EI_MAX_OVERFLOW_BUFFER_COUNT in tflite-model/tflite_learn_**_compiled.cpp. You MUST modify it in that model file, simply adding defines to main.cpp or Arduino sketch will not work. The exact number would require some trial and error / guesswork. Longer explanation (to let you know why switching offEI_CLASSIFIER_TFLITE_ENABLE_ESP_NN works, but would give you much slower inference): So, the issue here …

Try 2x kTensorArenaSize and make sure you are rebuilding the project properly (remove the build cache).