ref: Update and change several files

config.py

@@ -8,7 +8,7 @@
     'variance': [0.1, 0.2],
     'clip': False,
     'loc_weight': 2.0,
-    'milestones': [190, 220],
-    'epochs': 250,
+    'milestones': [100, 200],
+    'epochs': 300,
     'rgb_mean': (104, 117, 123)  # bgr order
 }

train.py

@@ -6,12 +6,11 @@
 import torch
 import torch.nn as nn
 from torch.utils.data import DataLoader
-from torch.optim.lr_scheduler import PolynomialLR, StepLR
 
 from config import cfg
 from models import FaceBoxes
 from layers import PriorBox, MultiBoxLoss
-from utils import Augmentation, VOCDetection, WiderFaceDetection
+from utils import Augmentation, WiderFaceDetection
 
 
 def parse_args():
@@ -22,7 +21,7 @@ def parse_args():
     # Dataset and data handling arguments
     parser.add_argument(
         '--train-data',
-        default='./data/widerface/train',
+        default='./data/WIDER_FACE/',
         type=str,
         help='Path to the training dataset directory.'
     )
@@ -170,16 +169,8 @@ def main(params):
     )
 
     # Learning rate scheduler
-    # lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=100, gamma=0.1)
     lr_scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=cfg['milestones'], gamma=params.gamma)
 
-    # iters_per_epoch = len(data_loader)
-    # lr_scheduler = PolynomialLR(
-    #     optimizer,
-    #     total_iters=iters_per_epoch * (params.epochs - params.lr_warmup_epochs),
-    #     power=params.power
-    # )
-
     start_epoch = 0
     if params.resume:
         try:
@@ -218,7 +209,7 @@ def main(params):
 
     #  save final model
     state = model.state_dict()
-    torch.save(state, f'{params.save_dir}/final.pth')
+    torch.save(state, f'{params.save_dir}/faceboxes.pth')
 
 
 if __name__ == '__main__':

utils/__init__.py

@@ -1,2 +1,2 @@
 from utils.transform import Augmentation
-from utils.dataset import VOCDetection, WiderFaceDetection
+from utils.dataset import  WiderFaceDetection

utils/dataset.py

@@ -1,3 +1,4 @@
+import torch.utils.data as data
 import os
 import cv2
 import numpy as np
@@ -40,7 +41,7 @@ def __call__(self, target):
         return bboxes_array
 
 
-class VOCDetection(Dataset):
+class WiderFaceDetection(Dataset):
 
     def __init__(self, root: str, transform=None):
         self.root = root
@@ -95,90 +96,3 @@ def collate_fn(batch):
             targets.append(torch.from_numpy(target).float())
 
         return torch.stack(images, 0), targets
-
-
-
-import os
-import cv2
-import numpy as np
-
-import torch
-import torch.utils.data as data
-
-
-class WiderFaceDetection(data.Dataset):
-    def __init__(self, root: str, transform=None):
-        self.root = root
-        self.transform = transform
-        self.image_paths = []
-        self.words = []
-        self._parse_labels(self.root)
-
-    def _parse_labels(self, root):
-        with open(os.path.join(root, 'label.txt'), 'r') as f:
-            lines = f.read().splitlines()
-
-        labels = []
-        for line in lines:
-            if line.startswith('#'):
-                if labels:
-                    self.words.append(labels.copy())
-                    labels.clear()
-                image_path = os.path.join(root, 'images', line[2:])
-                self.image_paths.append(image_path)
-            else:
-                labels.append([float(x) for x in line.split(' ')])
-        self.words.append(labels)
-
-    def __len__(self):
-        return len(self.image_paths)
-
-    def __getitem__(self, index):
-        image = cv2.imread(self.image_paths[index])
-        height, width, _ = image.shape
-
-        labels = self.words[index]
-        annotations = np.zeros((0, 5))
-        if not labels:
-            return torch.from_numpy(image), annotations
-
-        for label in labels:
-            annotation = np.zeros((1, 5))
-            # bbox
-            annotation[0, 0] = label[0]  # x1
-            annotation[0, 1] = label[1]  # y1
-            annotation[0, 2] = label[0] + label[2]  # x2
-            annotation[0, 3] = label[1] + label[3]  # y2
-            annotation[0,  4] = 1 # class idx
-
-            # landmarks
-            # annotation[0, 4] = label[4]    # l0_x
-            # annotation[0, 5] = label[5]    # l0_y
-            # annotation[0, 6] = label[7]    # l1_x
-            # annotation[0, 7] = label[8]    # l1_y
-            # annotation[0, 8] = label[10]   # l2_x
-            # annotation[0, 9] = label[11]   # l2_y
-            # annotation[0, 10] = label[13]  # l3_x
-            # annotation[0, 11] = label[14]  # l3_y
-            # annotation[0, 12] = label[16]  # l4_x
-            # annotation[0, 13] = label[17]  # l4_y
-            # annotation[0, 14] = 1 if label[4] >= 0 else -1
-
-            annotations = np.append(annotations, annotation, axis=0)
-
-        target = np.array(annotations)
-        if self.transform is not None:
-            image, target = self.transform(image, target)
-
-        return torch.from_numpy(image), target
-
-    @staticmethod
-    def collate_fn(batch):
-        images = []
-        targets = []
-        # Iterate over each data sample in the batch
-        for image, target in batch:
-            images.append(image)
-            targets.append(torch.from_numpy(target).float())
-
-        return torch.stack(images, 0), targets

webcam_inference.py

@@ -19,15 +19,15 @@ def parse_arguments():
     parser.add_argument(
         '-w', '--weights',
         type=str,
-        default='./weights/Resnet34_Final.pth',
+        default='./weights/faceboxes.pth',
         help='Path to the trained model weights'
     )
 
     # Detection settings
     parser.add_argument(
         '--conf-threshold',
         type=float,
-        default=0.4,
+        default=0.7,
         help='Confidence threshold for filtering detections'
     )
     parser.add_argument(