Updated the yolo model and ultralytics

app.py

@@ -29,7 +29,7 @@ except Exception as e:
     ndvi_model = None
 
 try:
-    yolo_model = load_yolo_model("4c_6c_regression.pt")
+    yolo_model = load_yolo_model("best_yolo_model.pt")
     logger.info("YOLO model loaded successfully")
 except Exception as e:
     logger.error(f"Failed to load YOLO model: {e}")

4c_6c_regression.pt → best_yolo_model.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:158f2a49f42e9e3ef40c12de1d86ddff9465f834cb8c006bb38a3fd6a8f99b15
-size 5998089
+oid sha256:5be2b24cd635b1e8f6e6b56f0ec49ba7751f627439670597807cec7e6f6fdf3e
+size 45153455

ultralytics/data/__pycache__/augment.cpython-312.pyc

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e072113c34b0a95bcb988e80f9d25235235c73fdb7319f0221e197c4a50779d9
+oid sha256:9f7f6b212a14a5068a553d00be673361718d528f2e6b2e324ea2a2e113a00191
 size 139583

ultralytics/data/augment.py

@@ -1541,7 +1541,12 @@ class LetterBox:
         self.stride = stride
         self.center = center  # Put the image in the middle or top-left
 
-    def __call__(self, labels=None, image=None):
+    def __call__(self, labels=None, image=None): # gets image during inference (1544, 2064, 4) but not during validation
+        # during validation, labels["img"].shape = (479, 640, 4)
+        # function returns labels with labels["img"].shape = (512, 672, 4)
+
+        # during inference the image is already provided, image.shape = (1544, 2064, 4)
+        # returns image with image.shape = (480, 640, 4)
         """
         Resizes and pads an image for object detection, instance segmentation, or pose estimation tasks.
 

ultralytics/data/base.py

@@ -160,8 +160,10 @@ class BaseDataset(Dataset):
                     LOGGER.warning(f"{self.prefix}WARNING ⚠️ Removing corrupt *.npy image file {fn} due to: {e}")
                     Path(fn).unlink(missing_ok=True)
                     im = cv2.imread(f, cv2.IMREAD_UNCHANGED)  # Load with all channels, including NDVI if present
+                    # assert 0, '=========== Read unchanged image ============='
             else:  # read image
                 im = cv2.imread(f, cv2.IMREAD_UNCHANGED)  # Load with all channels, including NDVI if present
+                # assert 0, '=========== Read unchanged image ============='
             if im is None:
                 raise FileNotFoundError(f"Image Not Found {f}")
 
@@ -208,6 +210,7 @@ class BaseDataset(Dataset):
         f = self.npy_files[i]
         if not f.exists():
             np.save(f.as_posix(), cv2.imread(self.im_files[i], cv2.IMREAD_UNCHANGED), allow_pickle=False)
+            # assert 0, '=========== Read unchanged image ============='
 
 
     def check_cache_disk(self, safety_margin=0.5):
@@ -219,6 +222,7 @@ class BaseDataset(Dataset):
         for _ in range(n):
             im_file = random.choice(self.im_files)
             im = cv2.imread(im_file, cv2.IMREAD_UNCHANGED)  # Load with all channels, including NDVI if present
+            # assert 0, '=========== Read unchanged image ============='
             if im is None:
                 continue
             b += im.nbytes
@@ -244,6 +248,7 @@ class BaseDataset(Dataset):
         n = min(self.ni, 30)  # extrapolate from 30 random images
         for _ in range(n):
             im = cv2.imread(random.choice(self.im_files), cv2.IMREAD_UNCHANGED)  # sample image
+            # assert 0, '=========== Read unchanged image ============='
             if im is None:
                 continue
             ratio = self.imgsz / max(im.shape[0], im.shape[1])  # max(h, w)  # ratio

ultralytics/data/dataset.py

@@ -560,12 +560,15 @@ class ClassificationDataset:
         if self.cache_ram:
             if im is None:  # Warning: two separate if statements required here, do not combine this with previous line
                 im = self.samples[i][3] = cv2.imread(f, cv2.IMREAD_UNCHANGED)
+                # assert 0, '=========== Read unchanged image ============='
         elif self.cache_disk:
             if not fn.exists():  # load npy
                 np.save(fn.as_posix(), cv2.imread(f, cv2.IMREAD_UNCHANGED), allow_pickle=False)
+                # assert 0, '=========== Read unchanged image ============='
             im = np.load(fn)
         else:  # read image
             im = cv2.imread(f, cv2.IMREAD_UNCHANGED)  # BGR
+            # assert 0, '=========== Read unchanged image ============='
         # Convert NumPy array to PIL image
         im = Image.fromarray(cv2.cvtColor(im, cv2.COLOR_BGR2RGB))
         sample = self.torch_transforms(im)

ultralytics/data/loaders.py

@@ -585,14 +585,37 @@ class LoadTensor:
         return self.bs
 
 
-def autocast_list(source):
+def autocast_list(source): # this function (as per experimentation) isn't called during validation or training, just inference
     """Merges a list of sources into a list of numpy arrays or PIL images for Ultralytics prediction."""
     files = []
     # assert 0, f'in def autocast_list() source: {source}'
+    # import itertools
     for im in source:
         if isinstance(im, (str, Path)):  # filename or uri
             # assert 0, f'im -> isinstance(im, (str, Path)) = True im=  {im}'
-            files.append(Image.open(requests.get(im, stream=True).raw if str(im).startswith("http") else im))
+            # assert 0, 'is the same func used in validation?'
+            # if str(im).startswith("http"):
+            # files.append(Image.open(requests.get(im, stream=True).raw if str(im).startswith("http") else im))
+
+            # pilimg = Image.open(requests.get(im, stream=True).raw if str(im).startswith("http") else im)
+            # orig_np_im = np.array(pilimg)
+            
+            # for perm in itertools.permutations([0, 1, 2, 3]):
+            #     np_im = orig_np_im[..., perm]
+            #     # np_im = np.ascontiguousarray(np_im)
+            #     files.append(Image.fromarray(np_im))
+
+            # np_im = orig_np_im[..., (0, 3, 2, 1)]
+            # np_im = np.ascontiguousarray(np_im)
+            # files.append(np_im)
+            if str(im).startswith("http"):
+                img = Image.open(requests.get(im, stream=True).raw)
+            else:
+                img = cv2.imread(im, cv2.IMREAD_UNCHANGED)
+                img = np.array(img)
+                img = img[..., (3,2,1,0)]
+                img = np.ascontiguousarray(img)
+            files.append(img)
         elif isinstance(im, (Image.Image, np.ndarray)):  # PIL or np Image
             files.append(im)
         else:
@@ -605,7 +628,7 @@ def autocast_list(source):
     # import numpy as np
     # np_arr = np.array(files[0])
     # print(f'np_arr.shape: {np_arr.shape}')
-    # assert 0, 'bye bye sucker'
+    # assert 0, 'bye bye'
     return files
 
 

ultralytics/engine/__pycache__/exporter.cpython-312.pyc

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d43eee4a96cf869121d1bc528ae7c8eed58d0f81baea7930b416f4a122dcf549
+oid sha256:b45eb6f61fa73af51b7cc3a8f73f05a4692f429d0d123cacdc8ee2af11704d11
 size 101374

ultralytics/engine/model.py

@@ -639,7 +639,7 @@ class Model(torch.nn.Module):
         args = {**self.overrides, **custom, **kwargs, "mode": "val"}  # highest priority args on the right
 
         validator = (validator or self._smart_load("validator"))(args=args, _callbacks=self.callbacks)
-        validator(model=self.model)
+        validator(model=self.model) # everything during validation happens here
         self.metrics = validator.metrics
         return validator.metrics
 

ultralytics/engine/predictor.py

@@ -117,7 +117,7 @@ class BasePredictor:
         self._lock = threading.Lock()  # for automatic thread-safe inference
         callbacks.add_integration_callbacks(self)
 
-    def preprocess(self, im):
+    def preprocess(self, im): # this preprocess is called during inference
         # assert 0, 'def preprocess() called in predictory.py file'
         # the line above confirmed that this function is not called during
         # model.train() or model.val()
@@ -140,9 +140,37 @@ class BasePredictor:
             
             im = np.stack(self.pre_transform(im))
             print(f'im.shape : {im.shape} in def preprocess() predictor.py after self.pre_transform()')
-            # im = im[..., ::-1].transpose((0, 3, 1, 2))  # BGR to RGB, BHWC to BCHW, (n, 3, h, w)
-            im = im[..., [2, 1, 0, 3]].transpose((0, 3, 1, 2)) # adjusted to handle NDVI band. [B, G, R, NDVI] to [R, G, B, NDVI]
-            # the transpose formats    BHWC  to   BCHW 
+            # im = im[..., ::-1].transpose((0, 3, 1, 2))  # BGR to RGB, BHWC to BCHW, (n, 3, h, w) 
+            # im = im.transpose((0, 3, 1, 2))  # BGR to RGB, BHWC to BCHW, (n, 3, h, w) 
+            
+            im = im.transpose((0, 3, 1, 2)) # adjusted to handle NDVI band. [B, G, R, NDVI] to [R, G, B, NDVI]
+            # im = im[..., [0, 1, 2, 3]].transpose((0, 3, 1, 2)) # adjusted to handle NDVI band. [B, G, R, NDVI] to [R, G, B, NDVI]
+            
+            # im = im[..., [0, 1, 2, 3]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [0, 1, 3, 2]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [0, 2, 1, 3]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [0, 2, 3, 1]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [0, 3, 1, 2]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [0, 3, 2, 1]].transpose((0, 3, 1, 2)) # ✅❌
+            # im = im[..., [1, 0, 2, 3]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [1, 0, 3, 2]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [1, 2, 0, 3]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [1, 2, 3, 0]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [1, 3, 0, 2]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [1, 3, 2, 0]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [2, 0, 1, 3]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [2, 0, 3, 1]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [2, 1, 0, 3]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [2, 1, 3, 0]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [2, 3, 0, 1]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [2, 3, 1, 0]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [3, 0, 1, 2]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [3, 0, 2, 1]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [3, 1, 0, 2]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [3, 1, 2, 0]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [3, 2, 0, 1]].transpose((0, 3, 1, 2)) # ❌
+            # im = im[..., [3, 2, 1, 0]].transpose((0, 3, 1, 2)) # ❌
+
             im = np.ascontiguousarray(im)  # contiguous
             # print(f'im.shape: {im.shape} in def preprocess() predictor.py')
             im = torch.from_numpy(im)
@@ -276,6 +304,22 @@ class BasePredictor:
                     # print(f'l276 pred.py im0s[0].shape: {im0s[0].shape}')
                     im = self.preprocess(im0s)
 
+                # preds[0].shape
+                # torch.Size([1, 43, 6300])
+
+                # preds[1][0][0].shape
+                # torch.Size([1, 71, 60, 80])
+                # preds[1][0][1].shape
+                # torch.Size([1, 71, 30, 40])
+                # preds[1][0][2].shape
+                # torch.Size([1, 71, 15, 20])
+
+                # preds[1][1][0].shape
+                # torch.Size([32, 6300])
+                
+                # preds[1][2][0].shape
+                # torch.Size([32, 120, 160])
+                
                 # Inference
                 with profilers[1]:
                     preds = self.inference(im, *args, **kwargs)

ultralytics/engine/validator.py

@@ -177,7 +177,7 @@ class BaseValidator:
             self.batch_i = batch_i
             # Preprocess
             with dt[0]:
-                batch = self.preprocess(batch)
+                batch = self.preprocess(batch) # preprocessing the validation data, models hasn't done anything yet
 
             # Inference
             with dt[1]:
@@ -190,7 +190,8 @@ class BaseValidator:
 
             # Postprocess
             with dt[3]:
-                preds = self.postprocess(preds)
+                preds = self.postprocess(preds) # in validation, preds does not contain the correct
+                # length of 39 per box, 4*coords, growth_val, 6*conf per each class, class (from 0-5, owing to 6 classes), 32*masks 
 
             self.update_metrics(preds, batch)
             if self.args.plots and batch_i < 3: