Model seems not to load the pretrained and low quality results

[theroguecode]

Thank you for the great work, the paper is very well presented and the method seems very novel and able to produce great results. I have run inference on a custom set of images using methods and functions from the code, but resulted in wrong outputs.
The mmcv api seems to have changed and was gives warnings: "gedepth/depth/models/backbones/depthformer_swin.py:915: UserWarning: DeprecationWarning: pretrained is a deprecated, please use "init_cfg" instead warnings.warn('DeprecationWarning: pretrained is a deprecated"
It appears that I was not loading the pretrained files for Kitti or DDAD that were downloaded from the repository.
After using the proper way in the config file only, according to open-mmlab/mmdetection#5370, it still gives a bad output, not resembling any depth but rather random. Is there some other place to modify the loading of the checkpoint? perhaps the Swin has to load its own weights first or the loaded Kitti/DDAD goes from the config to the builder in a way that it loads in the old api method?