Resume with custom anchors fix (#2361)

* Resume with custom anchors fix

* Update train.py

train.py

@@ -75,10 +75,8 @@ def train(hyp, opt, device, tb_writer=None, wandb=None):
         with torch_distributed_zero_first(rank):
             attempt_download(weights)  # download if not found locally
         ckpt = torch.load(weights, map_location=device)  # load checkpoint
-        if hyp.get('anchors'):
-            ckpt['model'].yaml['anchors'] = round(hyp['anchors'])  # force autoanchor
-        model = Model(opt.cfg or ckpt['model'].yaml, ch=3, nc=nc).to(device)  # create
-        exclude = ['anchor'] if opt.cfg or hyp.get('anchors') else []  # exclude keys
+        model = Model(opt.cfg or ckpt['model'].yaml, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device)  # create
+        exclude = ['anchor'] if (opt.cfg or hyp.get('anchors')) and not opt.resume else []  # exclude keys
         state_dict = ckpt['model'].float().state_dict()  # to FP32
         state_dict = intersect_dicts(state_dict, model.state_dict(), exclude=exclude)  # intersect
         model.load_state_dict(state_dict, strict=False)  # load
@@ -216,6 +214,7 @@ def train(hyp, opt, device, tb_writer=None, wandb=None):
             # Anchors
             if not opt.noautoanchor:
                 check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz)
+            model.half().float()  # pre-reduce anchor precision 
 
     # Model parameters
     hyp['box'] *= 3. / nl  # scale to layers