Segmentation model training issue and incorrect predictions

[aditidankar]

Description:
I followed the instructions provided in the README file to train the segmentation model, but I'm encountering a couple of issues. Firstly, the training metrics do not show any improvement over the epochs, indicating that the model is not learning properly. Secondly, when I attempted to visualize the outputs, I noticed that all predictions are being classified as the background class.

Steps to reproduce:

• Followed the instructions in the README file to set up the environment and dataset.
  

• Executed the training script, train_segmentation.py, with the specified parameters in pscn_seg.yml.
  

• Monitored the training metrics and visualized the outputs.
  

Expected behavior:

• The training metrics should show improvement over the epochs, indicating that the model is learning and converging.
  

• The visualized outputs should include accurate predictions, distinguishing different classes instead of solely predicting the background class.
  

Actual behavior:

• The training metrics did not exhibit any noticeable improvement over the epochs.
  

• All visualized outputs were predicted as the background class, regardless of the input.
  

Additional information:

• I verified that the dataset is correctly loaded and preprocessed.
  

• I checked the model architecture and the loss function, which appear to be set up correctly.
  

• I examined the training script for any potential issues but couldn't identify any obvious problems.
  

Environment:

• Operating System: Ubuntu 18.04.3
  

• Python version: 3.11.3
  

• GPU/CPU: NVIDIA GeForce RTX 2080 Ti
  

Any guidance or suggestions regarding this issue would be greatly appreciated. Thank you!