aded training script

data.yaml

@@ -0,0 +1,13 @@
+# Train/val/test sets as 
+## 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]
+
+
+path: /home/xdobro23/data/yolov8_16bit/  # dataset root dir
+train: images/train  # train images (relative to 'path') 128 images
+val: images/val # val images (relative to 'path') 128 images
+# test:  # test images (optional)
+
+# Classes
+nc: 1  # number of classes
+names: ['green']  # class names
+ch: 1

train.py

@@ -0,0 +1,6 @@
+from ultralytics import YOLO
+
+# Load a model
+model = YOLO('yolov8s.yaml')
+
+results = model.train(data='data.yaml', save_period=10, epochs=5000, batch=10, imgsz=1504, patience=100, device=1)

ultralytics/data/augment.py

@@ -163,7 +163,7 @@ class Mosaic(BaseMixTransform):
 
             # Place img in img4
             if i == 0:  # top left
-                img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8)  # base image with 4 tiles
+                img4 = np.full((s * 2, s * 2, 1), 114, dtype=np.float32)  # base image with 4 tiles
                 x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc  # xmin, ymin, xmax, ymax (large image)
                 x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h  # xmin, ymin, xmax, ymax (small image)
             elif i == 1:  # top right
@@ -176,7 +176,7 @@ class Mosaic(BaseMixTransform):
                 x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h)
                 x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)
 
-            img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b]  # img4[ymin:ymax, xmin:xmax]
+            img4[y1a:y2a, x1a:x2a, 0] = img[y1b:y2b, x1b:x2b]  # img4[ymin:ymax, xmin:xmax]
             padw = x1a - x1b
             padh = y1a - y1b
 
@@ -199,7 +199,7 @@ class Mosaic(BaseMixTransform):
 
             # Place img in img9
             if i == 0:  # center
-                img9 = np.full((s * 3, s * 3, img.shape[2]), 114, dtype=np.uint8)  # base image with 4 tiles
+                img9 = np.full((s * 3, s * 3, 1), 114, dtype=np.float32)  # base image with 4 tiles
                 h0, w0 = h, w
                 c = s, s, s + w, s + h  # xmin, ymin, xmax, ymax (base) coordinates
             elif i == 1:  # top
@@ -223,7 +223,7 @@ class Mosaic(BaseMixTransform):
             x1, y1, x2, y2 = (max(x, 0) for x in c)  # allocate coords
 
             # Image
-            img9[y1:y2, x1:x2] = img[y1 - padh:, x1 - padw:]  # img9[ymin:ymax, xmin:xmax]
+            img9[y1:y2, x1:x2, 0] = img[y1 - padh:, x1 - padw:]  # img9[ymin:ymax, xmin:xmax]
             hp, wp = h, w  # height, width previous for next iteration
 
             # Labels assuming imgsz*2 mosaic size
@@ -485,6 +485,7 @@ class RandomHSV:
     def __call__(self, labels):
         """Applies random horizontal or vertical flip to an image with a given probability."""
         img = labels['img']
+        return labels  # TODO:
         if self.hgain or self.sgain or self.vgain:
             r = np.random.uniform(-1, 1, 3) * [self.hgain, self.sgain, self.vgain] + 1  # random gains
             hue, sat, val = cv2.split(cv2.cvtColor(img, cv2.COLOR_BGR2HSV))
@@ -712,7 +713,7 @@ class Format:
     def __call__(self, labels):
         """Return formatted image, classes, bounding boxes & keypoints to be used by 'collate_fn'."""
         img = labels.pop('img')
-        h, w = img.shape[:2]
+        h, w = img.shape[0], img.shape[1]
         cls = labels.pop('cls')
         instances = labels.pop('instances')
         instances.convert_bbox(format=self.bbox_format)
@@ -906,5 +907,5 @@ class ToTensor:
         im = np.ascontiguousarray(im.transpose((2, 0, 1))[::-1])  # HWC to CHW -> BGR to RGB -> contiguous
         im = torch.from_numpy(im)  # to torch
         im = im.half() if self.half else im.float()  # uint8 to fp16/32
-        im /= 255.0  # 0-255 to 0.0-1.0
+        im /= 65535
         return im

ultralytics/data/loaders.py

@@ -249,7 +249,8 @@ class LoadImages:
         else:
             # Read image
             self.count += 1
-            im0 = cv2.imread(path)  # BGR
+            print("Grayscale image loading...")
+            im0 = cv2.imread(path, cv2.IMREAD_GRAYSCALE)  # BGR
             if im0 is None:
                 raise FileNotFoundError(f'Image Not Found {path}')
             s = f'image {self.count}/{self.nf} {path}: '

ultralytics/engine/predictor.py

@@ -122,7 +122,8 @@ class BasePredictor:
         not_tensor = not isinstance(im, torch.Tensor)
         if not_tensor:
             im = np.stack(self.pre_transform(im))
-            im = im[..., ::-1].transpose((0, 3, 1, 2))  # BGR to RGB, BHWC to BCHW, (n, 3, h, w)
+            im = np.expand_dims(im, -1)
+            im = im[..., ::].transpose((0, 3, 1, 2))  # BGR to RGB, BHWC to BCHW, (n, 3, h, w)
             im = np.ascontiguousarray(im)  # contiguous
             im = torch.from_numpy(im)
 
@@ -236,7 +237,7 @@ class BasePredictor:
 
         # Warmup model
         if not self.done_warmup:
-            self.model.warmup(imgsz=(1 if self.model.pt or self.model.triton else self.dataset.bs, 3, *self.imgsz))
+            self.model.warmup(imgsz=(1 if self.model.pt or self.model.triton else self.dataset.bs, 1, *self.imgsz))
             self.done_warmup = True
 
         self.seen, self.windows, self.batch, profilers = 0, [], None, (ops.Profile(), ops.Profile(), ops.Profile())

ultralytics/engine/validator.py

@@ -157,7 +157,7 @@ class BaseValidator:
             self.dataloader = self.dataloader or self.get_dataloader(self.data.get(self.args.split), self.args.batch)
 
             model.eval()
-            model.warmup(imgsz=(1 if pt else self.args.batch, 3, imgsz, imgsz))  # warmup
+            model.warmup(imgsz=(1 if pt else self.args.batch, 1, imgsz, imgsz))  # warmup
 
         dt = Profile(), Profile(), Profile(), Profile()
         n_batches = len(self.dataloader)

ultralytics/models/yolo/detect/train.py

@@ -53,7 +53,7 @@ class DetectionTrainer(BaseTrainer):
 
     def preprocess_batch(self, batch):
         """Preprocesses a batch of images by scaling and converting to float."""
-        batch['img'] = batch['img'].to(self.device, non_blocking=True).float() / 255
+        batch['img'] = batch['img'].to(self.device, non_blocking=True).float() / 65535.0  # uint16 to float16
         return batch
 
     def set_model_attributes(self):

ultralytics/models/yolo/detect/val.py

@@ -44,7 +44,7 @@ class DetectionValidator(BaseValidator):
     def preprocess(self, batch):
         """Preprocesses batch of images for YOLO training."""
         batch['img'] = batch['img'].to(self.device, non_blocking=True)
-        batch['img'] = (batch['img'].half() if self.args.half else batch['img'].float()) / 255
+        batch['img'] = (batch['img'].half() if self.args.half else batch['img'].float()) / 65535.0  # uint16 to float16
         for k in ['batch_idx', 'cls', 'bboxes']:
             batch[k] = batch[k].to(self.device)
 

ultralytics/nn/autobackend.py

@@ -445,7 +445,7 @@ class AutoBackend(nn.Module):
          """
         return torch.tensor(x).to(self.device) if isinstance(x, np.ndarray) else x
 
-    def warmup(self, imgsz=(1, 3, 640, 640)):
+    def warmup(self, imgsz=(1, 1, 640, 640)):
         """
         Warm up the model by running one forward pass with a dummy input.
 

ultralytics/utils/checks.py

@@ -448,6 +448,7 @@ def check_amp(model):
     Returns:
         (bool): Returns True if the AMP functionality works correctly with YOLOv8 model, else False.
     """
+    return False
     device = next(model.parameters()).device  # get model device
     if device.type in ('cpu', 'mps'):
         return False  # AMP only used on CUDA devices