Add CLI tool for exporting ONNX models

[runrunrun1994]

Signed-off-by: runrunrun1994 [email protected]

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

Merging #175 (096e511) into master (dce0263) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #175   +/-   ##
=======================================
  Coverage   96.16%   96.16%           
=======================================
  Files          10       10           
  Lines         600      600           
=======================================
  Hits          577      577           
  Misses         23       23           

Flag	Coverage Δ
unittests	96.16% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update dce0263...096e511. Read the comment docs.

[itsnine]

@zhiqwang @runrunrun1994 hello, it probably would be good if you add a feature for exporting onnx custom models, not only pretrained on coco dataset. I guess you should add something like torch.load_state_dict. What do you think?

[zhiqwang]

@zhiqwang @runrunrun1994 hello, it probably would be good if you add a feature for exporting onnx custom models, not only pretrained on coco dataset. I guess you should add something like torch.load_state_dict. What do you think?

Hi @itsnine @runrunrun1994 , This proposal is very good, because the trainer of yolort is still in the experimental stage, we have recently added a new interface to directly load yolov5 trained models #167, we can use this interface to implement this feature you mentioned.

from yolort.models import yolov5s

# the following 'yolov5s.pt' is downloaded from https://github.com/ultralytics/yolov5/releases/download/v4.0/yolov5s.pt
args.checkpoint_path = 'yolov5s.pt'
model = yolov5s(num_classes=args.num_classes, score_thresh=args.score_thresh, ...)
model.load_from_yolov5(args.checkpoint_path)
model.eval()

img_path = 'test/assets/bus.jpg'
results = model.predict(img_path)

[zhiqwang]

Hi @runrunrun1994 , Thanks for your contributions here. I left some comments, let me know your thoughts about these.