Poor results using end2end_inference_bodymesh

[Gowan1998]

Hello,
thank you very much for this great project. I have done many tests (inference only). The results are very bad and I want to ask you if I am doing something wrong? I did exactly what is suggested in DEMO.md. Should I do something additional?
Here some poor results:

[kevinlin311tw]

Our model will not work well if the target is not at the center of the input image. It is very sensitive to shifting.

[kevinlin311tw]

In this research topic, people generally normalize the reconstruction target at the center of the input image using bounding box (either from GT, object detector, or pose estimation). It is used for both training and testing.

If your testing images are not normalized as we did in training, our model may not work well.

[ghost]

Hello kevin,
thank you very much for your answer. I thought that detection is already implemented.
That means I have to use a detector (e.g. Yolo) before I use the inference script.