Codex/continual learning v2#10

Closed

thinksyncs wants to merge 160 commits intomainfrom

codex/continual-learning-v2

Commits on Feb 4, 2026

Update beads config and ignore list
GitHub Copilot
committed
Implement TensorRT exporter and build script
GitHub Copilot
committed
Add PNG mask loading in validator
GitHub Copilot
committed
Load PNG masks in training dataset
GitHub Copilot
committed
Document PNG mask loading in train_minimal
GitHub Copilot
committed
Mark bbox format support complete
GitHub Copilot
committed
Add TTA transform tests
GitHub Copilot
committed
Add target checks to latency benchmark
GitHub Copilot
committed
Improve TRT builder input name
GitHub Copilot
committed
Add TRT parity wrapper tool
GitHub Copilot
committed
Add TRT workflow helper
GitHub Copilot
committed
Add MIM masking to train_minimal
GitHub Copilot
committed
Add staged offset/K schedule
GitHub Copilot
committed
Add MIM teacher distillation
GitHub Copilot
committed
Expose backbone SPPF config
GitHub Copilot
committed
Add TTT MIM runner
GitHub Copilot
committed
Add inference constraints gating
GitHub Copilot
committed

Commits on Feb 5, 2026

Log MIM ratio in training
GitHub Copilot
committed
Add hybrid encoder level fusion
GitHub Copilot
committed
Integrate TTA into export_predictions
GitHub Copilot
committed
Add TTA runner base interface
GitHub Copilot
committed
Initial plan
Copilot
committed
Fix linting errors by removing unused imports and configuring ruff

Copilot
and
thinksyncs
committed
Finish TTA tooling
GitHub Copilot
committed
Gate depth loss by masks
GitHub Copilot
committed
Merge pull request #2 from thinksyncs/copilot/fix-action-errors
thinksyncs
authored
Add MIM mask schedule
GitHub Copilot
committed
Add update filters for TTT params
GitHub Copilot
committed
Add TTT integration plan
GitHub Copilot
committed
Mark staged offsets/K done
GitHub Copilot
committed
Add staged matcher cost schedule
GitHub Copilot
committed
Mark SIM jitter integration done
GitHub Copilot
committed
Add training/inference/export docs
GitHub Copilot
committed
Add CI scenario smoke run
GitHub Copilot
committed
Add scenario report schema
GitHub Copilot
committed
Add scenario suite runner
GitHub Copilot
committed
Add low-FP gate helper
GitHub Copilot
committed
Mark constraints gating done
GitHub Copilot
committed
Add symmetry template verification
GitHub Copilot
committed
Add inference utilities
GitHub Copilot
committed
Add denoising target augmentation
GitHub Copilot
committed
Export ONNX after training by default
GitHub Copilot
committed
Fix lint indentation in train_minimal
GitHub Copilot
committed
Add yolozu train/test CLI
GitHub Copilot
committed
Add yolozu train/test CLI
GitHub Copilot
committed
Allow insecure SSL in coco128 fetch
GitHub Copilot
committed
Use insecure SSL in CI fetch
GitHub Copilot
committed
Add metrics defaults and plotting support
GitHub Copilot
committed
Support mask-only labels
GitHub Copilot
committed
Derive labels from masks
GitHub Copilot
committed
Add Pillow to requirements
GitHub Copilot
committed
Add default YAML settings
GitHub Copilot
committed
Add SGD and LR schedules
GitHub Copilot
committed
Add predictions schema and adapter contract
GitHub Copilot
committed
docs: refresh README/spec; clarify TTA/TTT; fix interface notes
GitHub Copilot
committed
beads: add TTT integration remaining-work issues
GitHub Copilot
committed
docs: clarify TTA is prediction-space transform
GitHub Copilot
committed
feat: add LoRA option to minimal trainer
GitHub Copilot
committed
feat: add TTT (--ttt) to export_predictions
GitHub Copilot
committed
chore: bd sync
GitHub Copilot
committed
docs: highlight TTT and features in README
GitHub Copilot
committed
beads: add high-priority follow-up features
GitHub Copilot
committed
chore: bd sync
GitHub Copilot
committed
docs: clarify intrinsics units and frames
GitHub Copilot
committed
feat: LBFGS calibrate depth scale and intrinsics delta
GitHub Copilot
committed
test: add PyTorch vs ONNXRuntime parity
GitHub Copilot
committed
beads: close YOLOZU-8wj
GitHub Copilot
committed
beads: sync
GitHub Copilot
committed
lint: fix torch typing refs
GitHub Copilot
committed
adapter: realistic rtdetr_pose preprocessing
GitHub Copilot
committed
beads: close YOLOZU-4t3
GitHub Copilot
committed
beads: sync
GitHub Copilot
committed
predictions: validate wrapped meta contract
GitHub Copilot
committed
beads: close YOLOZU-jmv
GitHub Copilot
committed
beads: sync
GitHub Copilot
committed
dataset: add quality validator
GitHub Copilot
committed
Initial plan
Copilot
committed
beads: close YOLOZU-eco
GitHub Copilot
committed
beads: sync
GitHub Copilot
committed
Merge pull request #3 from thinksyncs/copilot/fix-issue-with-functionality
thinksyncs
authored
reports: unify baseline output
GitHub Copilot
committed
beads: close YOLOZU-vra
GitHub Copilot
committed
beads: sync
GitHub Copilot
committed
Initial plan
Copilot
committed
lora: add conv1x1 support
GitHub Copilot
committed
beads: close YOLOZU-59r
GitHub Copilot
committed
beads: sync
GitHub Copilot
committed
ttt: add presets and compact reports
GitHub Copilot
committed
beads: close YOLOZU-780
GitHub Copilot
committed
beads: sync
GitHub Copilot
committed
Fix test failures: add onnxscript dependency and fix ONNX compatibility

Copilot
and
thinksyncs
committed

Commits on Feb 6, 2026

Commits on Feb 7, 2026

Merge origin/main into thinksyncs/main
akira
committed
feat(dataset): add VOC/ADE20K seg prepare tools
akira
committed
feat(seg): add contracts and mIoU evaluator
akira
committed
fix(lint): resolve ruff F821/F811
akira
committed
feat(intrinsics): support OpenCV camera_matrix
akira
committed
docs: refresh README
akira
committed

Commits on Feb 8, 2026

feat: continual learning runner + eval
akira
committed
Add instance-seg PNG eval tool
akira
committed
Add instance-seg demo + validator
akira
committed
Improve instance-seg eval diagnostics
akira
committed
Add COCO instance-seg prediction converter
akira
committed
Add yolozu eval-instance-seg subcommand
akira
committed
Add README ON/OFF comparison image
akira
committed
Add allow-rgb-masks demo asset
akira
committed
Replace README option images with plots
akira
committed
Remove unconvincing demo plots from README
akira
committed
Add TTT presets and subset dataset tool
akira
committed
TTT: restore norm buffers on rollback/reset
akira
committed
README: remove instance-seg overlay image
akira
committed
docs: remove unused instance-seg overlay asset
akira
committed
Keypoints: YOLO pose labels + PCK eval tool
akira
committed
chore(beads): close YOLOZU-a1e
akira
committed
feat(keypoints): add OKS mAP + parity + benchmark
akira
committed
chore(beads): close YOLOZU-9uj
akira
committed
feat(license): add dependency license report
akira
committed
chore(beads): close YOLOZU-3v6
akira
committed
docs(runpod): pin yolozu TRT tag + add compliance script
akira
committed
chore(beads): close YOLOZU-0md
akira
committed
chore(notice): use ToppyMicroServices OÜ
akira
committed
chore(beads): close YOLOZU-hhf
akira
committed
chore(beads): claim YOLOZU-aa9
akira
committed
tests: add CUDA smoke for rtdetr_pose export
akira
committed
chore: avoid beads state in CUDA test PR
akira
committed
chore: keep beads sync out of code branch
akira
committed
chore: revert beads JSONL from code PR
akira
committed
feat: add LoRA options + Hessian CLI smoke tests
akira
committed
Beads: close YOLOZU-g6s
akira
committed
fix: default lora-freeze-base off
akira
committed
feat(intrinsics): parse OpenCV-style K
akira
committed
fix: remove duplicate lora guard
akira
committed
chore(beads): close YOLOZU-aa9
akira
committed
feat: rtdetr_pose trt parity+benchmark runner
akira
committed
chore(beads): start YOLOZU-6lb
akira
committed

Commits on Feb 9, 2026

feat(ttt): safe presets by default
akira
committed
chore(beads): ignore sync staging dir
akira
committed
YOLOZU-n97: pose eval + replay controls for continual learning
akira
committed