run yolov3-tiny engine tensorrt with python detection

[SokPhanith]

Env

• GPU jetson Nano.
• OS, Jetpack 4.5.1
• Cuda version 10.2
• TensorRT version 7.2

About this repo

yolov3-tiny

Your problem

Sorry, When I build yolov3-tiny I run detection with ./yolov3-tiny -d ../sample. It's work . How can get example with python detection with yolov3-tiny.engine and also scaled-yolov4, yolov3, yolov3-spp, yolov4

[wang-xinyu]

You can refer to yolov5_trt.py to implement it by yourself.

[SokPhanith]

I try run it and edit some code but I don't know why I got index of class sometime > 100 or sometime got wrong

[wang-xinyu]

You have to debug it yourself, since you have modified it.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[vjsrinivas]

@wang-xinyu If it's any use, here are the following modifications I did to get the yolov5_trt.py file working with YOLOv3:

1. Change 6001 to 7001 on line 161
2. Modify the post_process function to:

   def post_process(self, output, origin_h, origin_w):
       num = int(output[0])
       # Reshape to a two dimentional ndarray
       pred = np.reshape(output[1:], (-1, 7))[:num, :]
       if pred.shape[0] > 0:
           pred[:,4] *= pred[:,6]
           pred = pred[:,:-1]
       # Do nms
       boxes = self.non_max_suppression(pred, origin_h, origin_w, conf_thres=CONF_THRESH, nms_thres=IOU_THRESHOLD)
       result_boxes = boxes[:, :4] if len(boxes) else np.array([])
       result_scores = boxes[:, 4] if len(boxes) else np.array([])
       result_classid = boxes[:, 5] if len(boxes) else np.array([])
       return result_boxes, result_scores, result_classid

   For anyone else wondering the reason to the modifications, the output of YOLOv3 seemed to contain the objectness and class scores, so it was 7 elements in each prediction rather than YOLOv5's expected 6 elements (x,y,cw,ch,cls_id,cls_conf).

[wang-xinyu]

@vjsrinivas Can you please raise PR to add a yolov3_trt.py?

[vjsrinivas]

I will submit a PR in a couple days if someone else doesn't get to it first.