search

luxonis / depthai-experiments

Experimental projects we've done with DepthAI.
MIT License
809 stars 362 forks source link

Running models with FP16 #421

Open InzamamAnwar opened 1 year ago

InzamamAnwar commented 1 year ago

I am trying to run Face Detection model with FP16.

I converted the face detection model face-detection-retail-0004.blob with -ip FP16 using the OpenVINO compile_tool. The model converted successfully. But when I try to run it on OAK-D-PoE, it gives me the following error;

[DetectionNetwork(4)] [error] Input tensor 'data' (0) exceeds available data range. Data size (270000B), tensor offset (0), size (540000B) - skipping inference

I tried to setup setFp16 property of ColorCamera but still gives me an error.

Is there a way to run it with FP16 or FP32? My goal is to get same predictions with the model as it gives when doing inference with OpenVINO IR model on host machine

InzamamAnwar commented 1 year ago

The purpose of experimenting with different precision level on OAK-D-PoE is to get similar results as compared to CPU