danielbogdoll/iad-weights-a2-residual

A2: residual head on a frozen UNO model (native route)

Residual instance head trained on top of frozen UNO (M1) checkpoints, two variants (A2, A2_beta0).

26 files, 1.31 GiB. Paths are relative to the root of the research code; index.json lists each file's size and sha256. Each folder keeps the run manifest (command, commit, configuration) and/or the Detectron2 config.yaml of its training.

Folder	Checkpoints	Run	Command (first attempt)
output/A2_residual_frozenM1/A2	11	A2_residual_frozenM1	uv run python train_residual.py --config-file configs/cityscapes/semantic-segmentation/swin/residual_A2.yaml --num-gpus 1 --run A2_residual_frozenM1 --resume
output/A2_residual_frozenM1/A2_beta0	11	A2_residual_frozenM1	uv run python train_residual.py --config-file configs/cityscapes/semantic-segmentation/swin/residual_A2_beta0.yaml --num-gpus 1 --run A2_residual_frozenM1 --resume

Restore into a checkout of the research code: uv run python scripts/ops/backup_weights.py pull --only a2-residual.