Use of custom pretrained model with depth network

MAbdel-Ghani

Hi everyone, I am new to the depthai api and deploying models on the Oak-D (: I have a question regarding custom pretrained models. If I where to use an EfficientDet custom pretrained model, can I still get access to the stereo depth information using the spatial detection network? Would I be able to use the Yolo or the Mobilenet spatial detection network to get the spatial data even if my object detection model is neither mobilenet nor yolo?

erik

Hello MAbdel-Ghani, welcome to the community 🙂! One option would be to decode the NN result on the device using Script node, and configure SpatialLocationCalculator node (SLC in short) from the Script node (by sending SpatialLocationCalculatorConfig messages to the SLC).
Another option would be to decode the result on the host (just like in our efficientDet demo), and calculate spatial coordinates on the host (demo here).
Btw, we might also look into supporting the EfficientDet on the device, so we would have a SpatialEfficientDetNetwork or something like that.
Thanks, Erik