Human detection model

**Question/Issue: I want to develop an person detection model, using ESP32 S2 and arducam OV5642. The lowest resolution of camera is about 320x240. I have been following GitHub - Dasch0/esp32-arducam-edge-impulse: Minimal example code for running an Edge Impulse image classification network with the ESP32, ArduCAM, and PlatformIO implementation. I update this resource as per my components, but here using esp32 s2, to expand and use PSRAM i cam across How do I use psram in esp32-s2-wrover? - Development Platforms - PlatformIO Community. Well I am using platformio to configure and upload the code on board. With all this and using heap_malloc buffer as image buffer. The prediction is always no object and I also the buffer overflows. Alternative I tried to store the image on flash memory and then feed that to classifier, it gives false prediction. Could someone guide me where am I going wrong. Or point if this hardware combination is wrong.
P.S I have tried to train model on edge impulse using various resolution, like 96x96, 64x64 and 48x48. I am using object detection FOMO model

**Project ID: 318247

**Context/Use case: Human detection, edge computing

Hi @ajadhav

Did you manage to overcome this?

the Arduino IDE option tools for the XIAO_ESP32S2, set the option PSRAM: “OPI PSRAM” to enabled.



Ya its working, but the model performance is very poor, is FOMO not suitable for human detection task, I have about 520 training images