Wrong Inference Results of demo with pretrained

[sugarbeet34]

I followed the instruction to create a virtual environment (Ubuntu 22.04 with 3090 gpu, cuda10.1, python3.8), and ran the simple_extractor.py with the given pretrained model to infer on the demo.png, but got wrong answer as the image below.

And then I tried to create a new environment with the same version as the colab demo (cuda 12.2, python 3.10), but still got wrong answers.
And it's so strange that everytime I run the inference, the result image is wrong but different. Could anyone please tell me how to solve this? 😔

[sugarbeet34]

I tried the three pretrained models but none got available results, could anyone help? ::(