Hi,
I'm trying to implement external inference models on the STM32MP257-DK for an object detection project. Currently, I'm using the base project provided by ST in the SDK and modifying a few files, such as:
The base structure remains mostly the same, with only the model being switched by updating the reference in stai_mpu_object_detection.cc — choosing between SSD MobileNet and YOLOv8n.
After creating this “generic” module, I cross-compiled and transferred the binaries to the board. When selecting SSD MobileNet, everything works perfectly. However, when switching to YOLOv8n, no successful inference occurs.
Does anyone have any documentation or guidance on how to implement external models on this board?
I also implemented a debug printer that logs the total number of inferences. Here's what I get:
SSD Mobilenet
raw detections: 1
idx = 0 class = 1 score = 0.969
(g_print) Object info: person (96.9%)
raw detections: 0
raw detections: 2
idx=0 class=1 score=0.949
idx=1 class=1 score=0.609
(g_print) Object info: person (94.9%)
YoloV8n
raw detections: 0
raw detections: 0
...
