feat: add detector training, realistic eval, and browser export pipeline - #16

Merged

tkgstrator merged 19 commits into

developfrom

docs/training-plan-w384-refresh

Jul 13, 2026

Commits on Jul 13, 2026

docs(training-plan): refresh for w384 sweep, prune implemented items

tkgstrator
and
claude
committed
9a1ffc8
docs: add w384 sweep analysis with convergence metrics

tkgstrator
and
claude
committed
99f617f
feat(datasets): add HFDetectorDataset and resolution docs

tkgstrator
and
claude
committed
fdd704e
feat(train): add detector training script with HF dataset support

tkgstrator
and
claude
committed
f2a2217
feat(datasets): enhance detector augmentation pipeline

tkgstrator
and
claude
committed
575d581
feat(eval): add realistic eval script for boardocr checkpoints

tkgstrator
and
claude
committed
8b5ba42
feat(detector): track best checkpoint by configurable evaluation metric

tkgstrator
and
claude
committed
0d1e3ae
feat(eval): add error handling and skip counter to eval_realistic

tkgstrator
and
claude
committed
a7e5890
feat(eval): add W&B integration to eval_realistic

tkgstrator
and
claude
committed
6accfff
fix(eval): inject wandb cf access headers via environment variable

tkgstrator
and
claude
committed
10b8cf0
refactor(eval): require per-backbone wandb runs for 1:1 training pairing

tkgstrator
and
claude
committed
46f5211
docs(detector): add status document for board detector implementation

tkgstrator
and
claude
committed
95515f2
feat(eval): add end-to-end realistic eval with detector-predicted crops

tkgstrator
and
claude
committed
80b814b
docs: document w384 realistic evaluation and finalize backbone selection

tkgstrator
and
claude
committed
652ab0c
feat(export): add board_ocr export and checkpoint-aware factories

tkgstrator
and
claude
committed
a2b5b78
docs: add browser WebGPU inference time outlook

tkgstrator
and
claude
committed
d86d94c
feat(export): add ONNX export and GitHub Release pipeline

tkgstrator
and
claude
committed
41e12be
chore(release): bump version to 0.3.4

tkgstrator
and
claude
committed
5762e6d
fix(lint): remove unused imports and empty f-string

tkgstrator
and
claude
committed
0b569fd

Navigation Menu

Uh oh!

feat: add detector training, realistic eval, and browser export pipeline - #16

feat: add detector training, realistic eval, and browser export pipeline#16
tkgstrator merged 19 commits into
developfrom
docs/training-plan-w384-refresh

Commits on Jul 13, 2026

docs(training-plan): refresh for w384 sweep, prune implemented items

docs: add w384 sweep analysis with convergence metrics

feat(datasets): add HFDetectorDataset and resolution docs

feat(train): add detector training script with HF dataset support

feat(datasets): enhance detector augmentation pipeline

feat(eval): add realistic eval script for boardocr checkpoints

feat(detector): track best checkpoint by configurable evaluation metric

feat(eval): add error handling and skip counter to eval_realistic

feat(eval): add W&B integration to eval_realistic

fix(eval): inject wandb cf access headers via environment variable

refactor(eval): require per-backbone wandb runs for 1:1 training pairing

docs(detector): add status document for board detector implementation

feat(eval): add end-to-end realistic eval with detector-predicted crops

docs: document w384 realistic evaluation and finalize backbone selection

feat(export): add board_ocr export and checkpoint-aware factories

docs: add browser WebGPU inference time outlook

feat(export): add ONNX export and GitHub Release pipeline

chore(release): bump version to 0.3.4

fix(lint): remove unused imports and empty f-string