Feng Wang
> does not work with torch2trt converted models. Any extra information such as torch2trt version, error log etc.
I have already reverted this pr in #1792. @michelpromonet Your contribution is saved in README now.
> Since the release of [pycocotools version 2.0.9](https://pypi.org/project/pycocotools/2.0.9/), the loadRes function now requires the 'info' field to be present in the result file. If missing, using COCOEvaluator.evaluate_prediction raises a KeyError:...
> > > Since the release of [pycocotools version 2.0.9](https://pypi.org/project/pycocotools/2.0.9/), the loadRes function now requires the 'info' field to be present in the result file. If missing, using COCOEvaluator.evaluate_prediction raises...
Seems your dataset is None object. Did `get_dataloader` in your Exp object work well?
In yolox training logic, image size is changed per 10 iters, which causes the gpu memory raising/falling.
Your exp file.
Your issue is unclear. Do you mean onnxruntime code in cpp?
Sorry I could not catch your point. Could you say it a bit clear? Anything wrong in `demo_postprocess`?