GenSeg-Baselines / code /scripts /gen_hires_manifest.py
MaybeRichard's picture
Upload folder using huggingface_hub
057ec4b verified
Raw
History Blame Contribute Delete
2.61 kB
"""Resolution-aligned RE-TRAIN manifest: the 4 fully-conv framework methods on the
6 resolution-mismatched datasets, at a per-dataset higher img_size (matching ~nnU-Net's
working resolution). Reuses each arch's original command, swaps in --img_size R +
smaller batch for the bigger inputs, injects CUDA_DEVICE_ORDER=PCI_BUS_ID (else jobs
land on busy L40s). 4 archs x 6 datasets x 3 seeds = 72 jobs."""
import json, glob, re, os
# (dataset, protocol, R, batch)
DS = [
("cvc_clinicdb", "official", 384, 16),
("kvasir_seg", "official", 512, 8),
("busi", "fold01", 512, 8),
("fives", "official", 768, 4),
("refuge2", "official", 768, 4),
("idridd_segmentation", "fold01", 768, 4),
]
ARCHS = ["unet", "unetpp", "deeplabv3plus", "attention_unet"] # SwinUNet/TransUNet are res-locked
base = {} # (dataset, arch) -> seed0 cmd
for p in sorted(glob.glob(os.path.expanduser("~/.aris_queue/runs/20260605T13*/manifest.json"))):
d = json.load(open(p)); js = d.get("jobs") or d.get("phases", [{}])[0].get("jobs", [])
for j in js:
c = j.get("cmd", "")
m = re.search(r"--dataset (\S+).*?--arch (\S+)", c)
if m and "--seed 0" in c:
base.setdefault((m.group(1), m.group(2)), c)
jobs, missing = [], []
for ds, proto, R, B in DS:
for a in ARCHS:
bc = base.get((ds, a))
if not bc:
missing.append((ds, a)); continue
for s in (0, 1, 2):
c = bc
c = c.replace("export CUDA_VISIBLE_DEVICES=${GPU}",
"export CUDA_DEVICE_ORDER=PCI_BUS_ID CUDA_VISIBLE_DEVICES=${GPU}")
c = c.replace("--batch_size 16", "--batch_size %d" % B)
c = c.replace("--seed 0", "--img_size %d --seed %d" % (R, s)) # hits train.py AND test.py
out = "results/baselines/%s_%s/%s/seed%d/metrics.json" % (ds, proto, a, s)
jobs.append({"id": "hr_%s_%s_s%d" % (ds, a, s), "cmd": c, "expected_output": out})
if missing:
raise SystemExit("missing base cmds: %s" % missing)
manifest = {"project": "baselines_hires", "cwd": "/home/wzhang/LSC/Code/NPJ", "conda": "seggen",
"ssh": "a100", "gpus": [4, 5], "jobs_per_gpu": 2, "max_parallel": 4,
"gpu_free_threshold_mib": 40000, "oom_retry": {"delay": 300, "max_attempts": 3},
"phases": [{"name": "hr", "depends_on": [], "jobs": jobs}]}
RUN = "hires_" + os.environ["TS"]
rd = os.path.expanduser("~/.aris_queue/runs/" + RUN)
os.makedirs(rd + "/logs", exist_ok=True)
json.dump(manifest, open(rd + "/manifest.json", "w"), indent=2)
print("RUN=" + RUN, "jobs=" + str(len(jobs)))