research: audit blur routing and specify kernel feasibility gates

test: evaluate bounded deblur and document quality rejection

feat: add isolated regional sharpening experiments and gates

docs: plan regional sharpening with structure preservation

fix: update displayed version for sharpening release

fix: sharpen source detail before upscaling

perf: defer UI restoration loading and separate source blend

fix: preserve source detail and color during enhancement

feat(p3): complete automated Kaggle GPU 2-iteration verification gate

feat: package isolated two-step Kaggle GPU verification

docs: record full experimental baseline launch

feat: simplify dataset review with conservative experiment splits

docs: record technical image audits without identity decisions

docs: record completed dataset scan and review handoff

feat: add resumable dataset review and frozen baseline gates

docs: plan data review and staged model improvement

fix: restore pipeline correctness and gate training on verified data

docs: document Task 41 scheduler fix and Version 19 deployment

fix(scheduler): patch BasicSR CosineAnnealingRestartLR out-of-bounds crash on resume

fix(pipeline): force zero distortion on pure quality, fix radio state sync, and extend training to 6000 iters

docs: document Task 40 Kaggle GPU training completion and v3 model integration

fix(app): auto-enable Real-ESRGAN super-resolution in Zero-Distortion mode

feat(clarity): implement Zero-Distortion Wink Ultra-HD engine, face identity guard, and odd-dimension reshape padding

test(pipeline): modernize Phase 7 A/B model and preset verification suite

feat(universal): implement universal non-face image restoration, Real-ESRGAN multi-model engine, and compat shim

fix(realesrgan): patch degradations import directly via filesystem walk before training

fix(realesrgan): patch torchvision functional_tensor import in basicsr degradations

fix(kaggle): make GPU verification non-blocking in Real-ESRGAN notebook

fix(realesrgan): auto-generate missing train.py and fix CUDA sm_60 compatibility for Kaggle GPU

feat(ui): set Pure Quality & Sharpness as default preset for non-destructive universal enhancement

feat(realesrgan): add Kaggle GPU super-resolution training pipeline and runner

feat(pipeline): expose all model versions in get_available_models and UI dropdown

fix(wink): ensure 2D parse_mask in apply_crystal_skin_smoothing

fix(kaggle): prune intermediate checkpoints and increase save_checkpoint_freq to prevent disk full

fix(kaggle): resolve CUDA OOM, KeyError tensor(32) in ONNX export, and checkpoint export path

fix: kaggle onnxscript version conflict, wrap export in try/except, always save .pth checkpoint; fix openvino DLL provider spam

feat: v3 ONNX support, ArcFace paths, CPU batch fix, kaggle runner, get_available_models UI

fix(dataset): robust synthetic fallbacks for dataset preparation

fix: auto-install sm_60 compatible PyTorch on Tesla P100

docs: record Task 38 Phase 7 & GPU speedup in handover.md

fix: resolve GPU training bottleneck (num_gpu=1, batch_size=4, prefetch_mode=cuda, workers=4)

feat(Phase 7): add dynamic AI Model Engine selection and 1-click Windows desktop launchers

feat(v2.9.0): deliver Ultimate Glamour & Aesthetics Suite (crystal skin, glossy lips, doll-eye, golden hour)

fix: add CUDA verification and onnxscript dependency for Kaggle cloud

fix: reset resume_state to null for clean cloud GPU training and fix export command

fix: ensure basicsr/version.py is tracked and generated properly

fix: resolve dataset preparation pathing for Kaggle cloud training

feat: trigger automated Kaggle Cloud GPU Stage III training

fix: update Kaggle API payload structure and verified live deployment

feat: add automated Kaggle Cloud GPU training pipeline and runner