支持后台运行并记录日志

fed_run.py

@@ -44,12 +44,12 @@ def fed_run():
     train_txt = cfg.get("train_txt", "")
     if not train_txt:
         ds_root = cfg.get("dataset_path", "")
-        guess = os.path.join(ds_root, "train.txt") if ds_root else ""
+        guess = os.path.join(ds_root, "train2017.txt") if ds_root else ""
         train_txt = guess
 
     if not train_txt or not os.path.exists(train_txt):
         raise FileNotFoundError(
-            f"train.txt not found. Provide --config with 'train_txt' or ensure '{train_txt}' exists."
+            f"train2017.txt not found. Provide --config with 'train_txt' or ensure '{train_txt}' exists."
         )
 
     split = divide_trainset(
@@ -76,7 +76,7 @@ def fed_run():
 
     # --- build server & optional validation set ---
     server = FedYoloServer(client_list=users, model_name=model_name, params=cfg)
-    valset = build_valset_if_available(cfg, params=cfg, args=args_cli)
+    valset = build_valset_if_available(cfg, params=cfg, args=args_cli, val_name="val2017")
     # valset is a Dataset class, not data loader
     if valset is not None:
         server.load_valset(valset)

fed_run.sh

@@ -1,2 +1,2 @@
 GPUS=$1
-python3 -m torch.distributed.run --nproc_per_node=$GPUS fed_run.py ${@:2}
+nohup python3 -m torch.distributed.run --nproc_per_node=$GPUS fed_run.py ${@:2} > train.log 2>&1 & disown