Problems with configuration file (train.yaml)

[Peregalli]

I encountered an issue while using the model where I noticed occasional inference problems due to the edges of the blocks. To address this, I attempted to introduce a new parameter to overlay the cells. Upon inspection of the inference.py script, I found that it retrieves parameters from a configuration file, such as cell_size. Although I couldn't pinpoint where it loads these parameters, I assumed it was from configs/train.yaml. Consequently, I added the following lines (in a similar style) to the script:

@property
    def overlap(self):
        return self._config.get("overlap")

Additionally, I included the parameter in the .yaml file:overlap: 0.5. However, the solution didn't resolve the issue. Furthermore, I observed that altering the cell_block parameter in the .yaml file didn't reflect in the debugging session, suggesting it might be loaded from elsewhere. Where are those parameters being loaded from?

[serykhelena]

Hello @Peregalli! Could you kindly provide more details? Are you attempting to run inferences on already trained models from the checkpoint folder, or is this a custom-trained model of your own? Additionally, could you please describe the steps to reproduce the inference issues you have mentioned?