Summary of changes:

• Moved the conditions: block from the data: section into the new inference: block in pytorch_config.yaml. This is more logical because the data: block should only configure data augmentation.

• Updated the net_type used in the CTD/BUCTD Colab notebook from ctd_prenet_rtmpose_m to ctd_coam_w32, as the ctd_prenet_rtmpose_m variant has shown issues during training. The fix for all ctd_prenet_rtmpose variants will be implemented in a future PR.

• Added documentation for the inference: block in the PyTorch configuration docs, covering all sub-configs:

  • multithreading – controls preprocessing/prediction threading during inference.
  • compile – optional torch.compile settings.
  • autocast – optional mixed-precision settings.
  • conditions – specifies conditions to use for Conditional Top-Down models.

Notes:

• These changes improve the clarity and maintainability of the configuration file by separating data augmentation from inference-specific settings.

• Default behaviors remain unchanged for all users.