fix: always set class_names from dataset annotations

[lixiaolei1982]

• Remove conditional check that only set class_names when None
• Unconditionally assign dataset-derived class_names to train_config
• Ensure consistency between dataset annotations and training configuration

Description

The current code in train_from_config method contains an incorrect conditional check:
if "class_names" in train_config and train_config["class_names"] is None:
train_config["class_names"] = class_names

This logic is flawed because:
1.It only sets class_names when it already exists in train_config AND is None
2.It doesn't handle the case where class_names might not exist in train_config at all
3.It doesn't override potentially incorrect class_names values that are non-None

The correct behavior should be to always set train_config["class_names"] to the class_names variable derived from the dataset (Roboflow or COCO), ensuring consistency between the dataset annotations and the training configuration.

#Changes:

Remove the conditional check and unconditionally set train_config["class_names"] = class_names
This ensures the class names from the actual dataset annotations are always used for training

#Impact:

Fixes potential mismatches between dataset classes and training configuration
Ensures consistent behavior regardless of initial train_config["class_names"] value
Maintains backward compatibility while fixing the logic error

[Copilot]

Pull request overview

This PR fixes a logic flaw in the train_from_config method where class names from the dataset annotations were only conditionally assigned to the training configuration. The change ensures that class names derived from the dataset are always used, maintaining consistency between dataset annotations and training configuration.

Changes:

• Remove conditional check that only set class_names when it existed and was None
• Unconditionally assign dataset-derived class_names to train_config

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Borda]

@lixiaolei1982, could you please elaborate on why the actual behaviour is wrong?