4-Channels compatibility [WorkInProgress]

data/hyp.scratch.s.yaml

@@ -2,7 +2,7 @@ lr0: 0.01  # initial learning rate (SGD=1E-2, Adam=1E-3)
 lrf: 0.1  # final OneCycleLR learning rate (lr0 * lrf)
 momentum: 0.937  # SGD momentum/Adam beta1
 weight_decay: 0.0005  # optimizer weight decay 5e-4
-warmup_epochs: 3.0  # warmup epochs (fractions ok)
+warmup_epochs: 0.0  # warmup epochs (fractions ok)
 warmup_momentum: 0.8  # warmup initial momentum
 warmup_bias_lr: 0.1  # warmup initial bias lr
 box: 0.05  # box loss gain

data/hyp.scratch.yaml

@@ -1,16 +1,16 @@
-lr0: 0.01  # initial learning rate (SGD=1E-2, Adam=1E-3)
-lrf: 0.1  # final OneCycleLR learning rate (lr0 * lrf)
-momentum: 0.937  # SGD momentum/Adam beta1
+lr0: 0.001  # initial learning rate (SGD=1E-2, Adam=1E-3)
+lrf: 0.01  # final OneCycleLR learning rate (lr0 * lrf)
+momentum: 0.80  # SGD momentum/Adam beta1
 weight_decay: 0.0005  # optimizer weight decay 5e-4
-warmup_epochs: 3.0  # warmup epochs (fractions ok)
+warmup_epochs: 4.0  # warmup epochs (fractions ok)
 warmup_momentum: 0.8  # warmup initial momentum
 warmup_bias_lr: 0.1  # warmup initial bias lr
-box: 0.05  # box loss gain
-cls: 0.3  # cls loss gain
+box: 0.006  # box loss gain
+cls: 0.006  # cls loss gain
 cls_pw: 1.0  # cls BCELoss positive_weight
-obj: 0.7  # obj loss gain (scale with pixels)
+obj: 0.07  # obj loss gain (scale with pixels)
 obj_pw: 1.0  # obj BCELoss positive_weight
-iou_t: 0.20  # IoU training threshold
+iou_t: 0.10  # IoU training threshold
 anchor_t: 4.0  # anchor-multiple threshold
 # anchors: 3  # anchors per output layer (0 to ignore)
 fl_gamma: 0.0  # focal loss gamma (efficientDet default gamma=1.5)
@@ -22,7 +22,7 @@ translate: 0.1  # image translation (+/- fraction)
 scale: 0.9  # image scale (+/- gain)
 shear: 0.0  # image shear (+/- deg)
 perspective: 0.0  # image perspective (+/- fraction), range 0-0.001
-flipud: 0.0  # image flip up-down (probability)
+flipud: 0.50  # image flip up-down (probability)
 fliplr: 0.5  # image flip left-right (probability)
 mosaic: 1.0  # image mosaic (probability)
-mixup: 0.0  # image mixup (probability)
+mixup: 0.5  # image mixup (probability)

models/models.py

@@ -11,14 +11,13 @@ def create_modules(module_defs, img_size, cfg):
 
     img_size = [img_size] * 2 if isinstance(img_size, int) else img_size  # expand if necessary
     _ = module_defs.pop(0)  # cfg training hyperparams (unused)
-    output_filters = [3]  # input channels
+    output_filters = [_["channels"]]  # input channels
     module_list = nn.ModuleList()
     routs = []  # list of layers which rout to deeper layers
     yolo_index = -1
 
     for i, mdef in enumerate(module_defs):
         modules = nn.Sequential()
-
         if mdef['type'] == 'convolutional':
             bn = mdef['batch_normalize']
             filters = mdef['filters']
@@ -90,7 +89,7 @@ def create_modules(module_defs, img_size, cfg):
                 modules.add_module('activation', Mish())
             elif mdef['activation'] == 'silu':
                 modules.add_module('activation', nn.SiLU())
-                
+
         elif mdef['type'] == 'dropout':
             p = mdef['probability']
             modules = nn.Dropout(p)
@@ -208,7 +207,7 @@ def create_modules(module_defs, img_size, cfg):
                 bias.data[:, 4] += math.log(8 / (640 / stride[yolo_index]) ** 2)  # obj (8 objects per 640 image)
                 bias.data[:, 5:] += math.log(0.6 / (modules.nc - 0.99))  # cls (sigmoid(p) = 1/nc)
                 module_list[j][0].bias = torch.nn.Parameter(bias_, requires_grad=bias_.requires_grad)
-                
+
                 #j = [-2, -5, -8]
                 #for sj in j:
                 #    bias_ = module_list[sj][0].bias
@@ -459,7 +458,6 @@ def __init__(self, cfg, img_size=(416, 416), verbose=False):
         self.info(verbose) if not ONNX_EXPORT else None  # print model description
 
     def forward(self, x, augment=False, verbose=False):
-
         if not augment:
             return self.forward_once(x)
         else:  # Augment images (inference and test only) https://github.com/ultralytics/yolov3/issues/931
@@ -518,8 +516,6 @@ def forward_once(self, x, augment=False, verbose=False):
             elif name == 'JDELayer':
                 yolo_out.append(module(x, out))
             else:  # run module directly, i.e. mtype = 'convolutional', 'upsample', 'maxpool', 'batchnorm2d' etc.
-                #print(module)
-                #print(x.shape)
                 x = module(x)
 
             out.append(x if self.routs[i] else [])

utils/datasets.py

@@ -924,7 +924,7 @@ def load_image(self, index):
     img = self.imgs[index]
     if img is None:  # not cached
         path = self.img_files[index]
-        img = cv2.imread(path)  # BGR
+        img = cv2.imread(path, cv2.IMREAD_UNCHANGED)  # BGR
         assert img is not None, 'Image Not Found ' + path
         h0, w0 = img.shape[:2]  # orig hw
         r = self.img_size / max(h0, w0)  # resize image to img_size

utils/parse_config.py

@@ -21,7 +21,7 @@ def parse_model_cfg(path):
             mdefs[-1]['type'] = line[1:-1].rstrip()
             if mdefs[-1]['type'] == 'convolutional':
                 mdefs[-1]['batch_normalize'] = 0  # pre-populate with zeros (may be overwritten later)
-        
+
         else:
             key, val = line.split("=")
             key = key.rstrip()

utils/plots.py

@@ -90,6 +90,7 @@ def output_to_target(output, width, height):
     # Convert model output to target format [batch_id, class_id, x, y, w, h, conf]
     if isinstance(output, torch.Tensor):
         output = output.cpu().numpy()
+    output = output.cpu().numpy()
 
     targets = []
     for i, o in enumerate(output):
@@ -137,11 +138,14 @@ def plot_images(images, targets, paths=None, fname='images.jpg', names=None, max
     for i, img in enumerate(images):
         if i == max_subplots:  # if last batch has fewer images than we expect
             break
-
         block_x = int(w * (i // ns))
         block_y = int(h * (i % ns))
 
         img = img.transpose(1, 2, 0)
+        #IF Image has 4 channels convert to plot
+        if img.shape[2] == 4:
+            img = cv2.cvtColor(img, cv2.COLOR_BGRA2RGB)
+
         if scale_factor < 1:
             img = cv2.resize(img, (w, h))