Different width and height for YoloV3

[adrian2]

Hi I'm trying to do inference with intel OpenVino I've tried following the tutorial at here and I'm stuck at the step where I need to convert it into Tensorflow pb. After some researching, I've found out that you need to change some code on the conversion at upsample layer based on this forum post and I did that, but now my model doesn't detect any object. Is there any solution to this problem?

P.S. I'm using Yolov3 with SPP for my configuration and followed the conversion script readme to convert with spp included in the model, my detection size is (32,224).

[SuryaPSC_Intel]

Hi adrian,

Converting YOLOv3 model to IR suggests to use yolov3.weights (for the YOLOv3 model) or yolov3-tiny.weights (for the YOLOv3-tiny model). Kindly try using the suggested models and weight files.

Best Regards,

Surya

[adrian2]

Chauhan, Surya Pratap Singh (Intel) wrote:

Hi adrian,

Converting YOLOv3 model to IR suggests to use yolov3.weights (for the YOLOv3 model) or yolov3-tiny.weights (for the YOLOv3-tiny model). Kindly try using the suggested models and weight files.

Best Regards,

Surya

Hi, thanks for the reply, I'm trying to convert my own custom object YOLO detection, and my model have been trained to detect in size of (64, 224) as opposed to the original YOLOv3 (416, 416). While I'm trying to convert it with the github link provided by the tutorial, It throws an error, so is there any workaround?