Could you please give us some explanation on how to run inference with trained model?

[Ali-Tohidifar]

I’ve developed a custom inference script to run predictions using BoundaryFormer. The training completed successfully, but I'm encountering errors during inference. Below is a summary of my approach and the problems I'm facing:

Inference Script Overview:

• I am using Detectron2 v0.6 for the inference script.
• The model and configuration are loaded using get_cfg() from Detectron2, and I run inference on custom dataset.
• I use DefaultPredictor from Detectron2 for the inference process.

Errors Encountered:

• Parameter Shape Mismatches:
  Multiple parameters are skipped due to shape mismatches between the model architecture and the saved checkpoint. For example:

Skip loading parameter 'proposal_generator.rpn_head.conv.weight' to the model due to incompatible shapes...
Skip loading parameter 'roi_heads.box_predictor.cls_score.weight'...

Questions:

1. What is the recommended way to run inference using a custom-trained BoundaryFormer model with Detectron2 v0.6?
2. Are the parameter mismatches during inference expected, or is there a way to ensure better compatibility between the model and checkpoint?

Any guidance on resolving these issues would be greatly appreciated. Thank you for your help!