feat(training): add gradual frequency-loss transition

[Haruko386]

Summary

Stage 2 previously changed directly from latent MSE + pixel L1 reconstruction to latent frequency loss after iteration 20,000. Add a boolean latent_freq_loss.gradual_transition setting so this transition can be linear instead of abrupt.

With the setting enabled in config/train_apdepth.yaml, the reconstruction weight decreases from 1 to 0 while the frequency-loss weight increases from 0 to 1 between iteration 20,000 and max_iter. Setting it to false, or omitting it in an older saved config, preserves the original hard switch. TensorBoard now records train/freq_loss_weight.

Type of Change

• Bug fix
• New feature
• Documentation update
• Refactor
• CI / build change
• Other

Test Results

• Verified disabled mode preserves the exact 20,000/20,001 hard-switch boundary.
• Verified enabled weights at iterations 20,000/20,250/20,500/20,750/21,000 are 0/0.25/0.5/0.75/1.
• Verified blended loss is finite and backpropagates finite gradients.
• Parsed all YAML files and ran Python compile checks.
• ruff check . --select E9,F63,F7,F82 and git diff --check passed.

[coderabbitai]

Warning

Review limit reached

Next included review available in 4 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Advanced

Run ID: dbe92f4e-6323-44bd-b668-8497504c1d2b

📥 Commits

Reviewing files that changed from the base of the PR and between 866708d and 1dae11e.

📒 Files selected for processing (3)

• README.md
• config/train_apdepth.yaml
• src/trainer/apdepth_trainer.py

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}