Add autobatch feature for best batch-size estimation (#5092)

* Autobatch

* fix mem

* fix mem2

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Update train.py

* print result

* Cleanup print result

* swap fix in call

* to 64

* use total

* fix

* fix

* fix

* fix

* fix

* Update

* Update

* Update

* Update

* Update

* Update

* Update

* Cleanup printing

* Update final printout

* Update autobatch.py

* Update autobatch.py

* Update autobatch.py

train.py

@@ -36,6 +36,7 @@ import val  # for end-of-epoch mAP
 from models.experimental import attempt_load
 from models.yolo import Model
 from utils.autoanchor import check_anchors
+from utils.autobatch import check_train_batch_size
 from utils.datasets import create_dataloader
 from utils.general import labels_to_class_weights, increment_path, labels_to_image_weights, init_seeds, \
     strip_optimizer, get_latest_run, check_dataset, check_git_status, check_img_size, check_requirements, \
@@ -131,6 +132,14 @@ def train(hyp,  # path/to/hyp.yaml or hyp dictionary
             print(f'freezing {k}')
             v.requires_grad = False
 
+    # Image size
+    gs = max(int(model.stride.max()), 32)  # grid size (max stride)
+    imgsz = check_img_size(opt.imgsz, gs, floor=gs * 2)  # verify imgsz is gs-multiple
+
+    # Batch size
+    if RANK == -1 and batch_size == -1:  # single-GPU only, estimate best batch size
+        batch_size = check_train_batch_size(model, imgsz)
+
     # Optimizer
     nbs = 64  # nominal batch size
     accumulate = max(round(nbs / batch_size), 1)  # accumulate loss before optimizing
@@ -190,11 +199,6 @@ def train(hyp,  # path/to/hyp.yaml or hyp dictionary
 
         del ckpt, csd
 
-    # Image sizes
-    gs = max(int(model.stride.max()), 32)  # grid size (max stride)
-    nl = model.model[-1].nl  # number of detection layers (used for scaling hyp['obj'])
-    imgsz = check_img_size(opt.imgsz, gs, floor=gs * 2)  # verify imgsz is gs-multiple
-
     # DP mode
     if cuda and RANK == -1 and torch.cuda.device_count() > 1:
         logging.warning('DP not recommended, instead use torch.distributed.run for best DDP Multi-GPU results.\n'
@@ -242,6 +246,7 @@ def train(hyp,  # path/to/hyp.yaml or hyp dictionary
         model = DDP(model, device_ids=[LOCAL_RANK], output_device=LOCAL_RANK)
 
     # Model parameters
+    nl = model.model[-1].nl  # number of detection layers (to scale hyps)
     hyp['box'] *= 3. / nl  # scale to layers
     hyp['cls'] *= nc / 80. * 3. / nl  # scale to classes and layers
     hyp['obj'] *= (imgsz / 640) ** 2 * 3. / nl  # scale to image size and layers
@@ -440,7 +445,7 @@ def parse_opt(known=False):
     parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='dataset.yaml path')
     parser.add_argument('--hyp', type=str, default=ROOT / 'data/hyps/hyp.scratch.yaml', help='hyperparameters path')
     parser.add_argument('--epochs', type=int, default=300)
-    parser.add_argument('--batch-size', type=int, default=16, help='total batch size for all GPUs')
+    parser.add_argument('--batch-size', type=int, default=16, help='total batch size for all GPUs, -1 for autobatch')
     parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='train, val image size (pixels)')
     parser.add_argument('--rect', action='store_true', help='rectangular training')
     parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')