This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

AM5729: TIDL Import problem

Part Number: AM5729
Other Parts Discussed in Thread: BEAGLEBONE

Hi,

I have a problem with TIDL Importer from the AM57XX SDK. Im trying to import an object detection model trained on Caffe-Jacinto. The model seems to be exported right because when it comes to calibration I can see this output

Target: number label value xmin  ymin  xmax  ymax
Target:  0.00  1.00  0.42 -0.05  0.22  0.99  0.86

So this means that the model detects the object on the image. But when I'm using this exported model on the BeagleBone AI in the SSD example, it doesn't detect anything. This same happens on the model on which my model was fine-tuned. I've used this model for fine-tuning.

https://github.com/mathmanu/caffe-jacinto-models/tree/caffe-0.17/trained/object_detection/voc0712/JDetNet/ssd512x512_ds_PSP_dsFac_32_fc_0_hdDS8_1_kerMbox_3_1stHdSameOpCh_1

The problem is not with the confidence threshold, because even when I set it to 1 doesn't change anything. So how should the model be imported to run on BB AI?