updates

test.py

@@ -148,7 +148,7 @@ def test(
 
     # Print results
     pf = '%20s' + '%10.3g' * 6  # print format
-    print(pf % ('all', seen, nt.sum(), mp, mr, map, mf1), end='\n\n')
+    print(pf % ('all', seen, nt.sum(), mp, mr, map, mf1))
 
     # Print results per class
     if nc > 1 and len(stats):

train.py

@@ -75,7 +75,7 @@ def train(
     device = torch_utils.select_device()
 
     if multi_scale:
-        img_size = 608  # initiate with maximum multi_scale size
+        img_size = round((img_size / 32) * 1.5) * 32  # initiate with maximum multi_scale size
         opt.num_workers = 0  # bug https://github.com/ultralytics/yolov3/issues/174
     else:
         torch.backends.cudnn.benchmark = True  # unsuitable for multiscale
@@ -138,7 +138,13 @@ def train(
     # plt.savefig('LR.png', dpi=300)
 
     # Dataset
-    dataset = LoadImagesAndLabels(train_path, img_size, batch_size, augment=True, rect=False, cache=True)
+    dataset = LoadImagesAndLabels(train_path,
+                                  img_size,
+                                  batch_size,
+                                  augment=True,
+                                  rect=False,
+                                  cache=True,
+                                  multi_scale=multi_scale)
 
     # Initialize distributed training
     if torch.cuda.device_count() > 1:

utils/datasets.py

@@ -130,14 +130,19 @@ class LoadWebcam:  # for inference
 
 
 class LoadImagesAndLabels(Dataset):  # for training/testing
-    def __init__(self, path, img_size=416, batch_size=16, augment=False, rect=True, image_weights=False, cache=False):
+    def __init__(self, path, img_size=416, batch_size=16, augment=False, rect=True, image_weights=False, cache=False,
+                 multi_scale=False):
         with open(path, 'r') as f:
             img_files = f.read().splitlines()
             self.img_files = list(filter(lambda x: len(x) > 0, img_files))
 
         n = len(self.img_files)
-        self.n = n
+        bi = np.floor(np.arange(n) / batch_size).astype(np.int)  # batch index
+        nb = bi[-1] + 1  # number of batches
         assert n > 0, 'No images found in %s' % path
+
+        self.n = n
+        self.batch = bi  # batch index of image
         self.img_size = img_size
         self.augment = augment
         self.image_weights = image_weights
@@ -148,11 +153,13 @@ class LoadImagesAndLabels(Dataset):  # for training/testing
                                 replace('.bmp', '.txt').
                                 replace('.png', '.txt') for x in self.img_files]
 
+        if multi_scale:
+            s = img_size / 32
+            self.multi_scale = ((np.linspace(0.5, 1.5, nb) * s).round().astype(np.int) * 32)
+
         # Rectangular Training  https://github.com/ultralytics/yolov3/issues/232
         if self.rect:
             from PIL import Image
-            bi = np.floor(np.arange(n) / batch_size).astype(np.int)  # batch index
-            nb = bi[-1] + 1  # number of batches
 
             # Read image shapes
             sp = 'data' + os.sep + path.replace('.txt', '.shapes').split(os.sep)[-1]  # shapefile path
@@ -182,7 +189,6 @@ class LoadImagesAndLabels(Dataset):  # for training/testing
                     shapes[i] = [1, 1 / mini]
 
             self.batch_shapes = np.ceil(np.array(shapes) * img_size / 32.).astype(np.int) * 32
-            self.batch = bi  # batch index of image
 
         # Preload images
         if cache and (n < 1001):  # preload all images into memory if possible
@@ -207,6 +213,12 @@ class LoadImagesAndLabels(Dataset):  # for training/testing
     def __len__(self):
         return len(self.img_files)
 
+    # def __iter__(self):
+    #     self.count = -1
+    #     print('ran dataset iter')
+    #     #self.shuffled_vector = np.random.permutation(self.nF) if self.augment else np.arange(self.nF)
+    #     return self
+
     def __getitem__(self, index):
         if self.image_weights:
             index = self.indices[index]
@@ -242,10 +254,11 @@ class LoadImagesAndLabels(Dataset):  # for training/testing
         # Letterbox
         h, w, _ = img.shape
         if self.rect:
-            new_shape = self.batch_shapes[self.batch[index]]
-            img, ratio, padw, padh = letterbox(img, new_shape=new_shape, mode='rect')
+            shape = self.batch_shapes[self.batch[index]]
+            img, ratio, padw, padh = letterbox(img, new_shape=shape, mode='rect')
         else:
-            img, ratio, padw, padh = letterbox(img, new_shape=self.img_size, mode='square')
+            shape = int(self.multi_scale[self.batch[index]]) if hasattr(self, 'multi_scale') else self.img_size
+            img, ratio, padw, padh = letterbox(img, new_shape=shape, mode='square')
 
         # Load labels
         labels = []

utils/gcp.sh

@@ -72,7 +72,7 @@ rm -rf darknet && git clone https://github.com/AlexeyAB/darknet && cd darknet &&
 ./darknet detector train ../supermarket2/supermarket2.data cfg/yolov3-spp-sm2-1cls.cfg darknet53.conv.74 -map -dont_show # train
 ./darknet detector train ../supermarket2/supermarket2.data cfg/yolov3-spp-sm2-1cls.cfg backup/yolov3-spp-sm2-1cls_last.weights  # resume
 python3 train.py --data ../supermarket2/supermarket2.data --cfg cfg/yolov3-spp-sm2-1cls.cfg  # test
-python3 test.py --data ../supermarket2/supermarket2.data --weights ../darknet/backup/yolov3-spp-sm2-1cls_3000.weights  # test
+python3 test.py --data ../supermarket2/supermarket2.data --weights ../darknet/backup/yolov3-spp-sm2-1cls_5000.weights --cfg cfg/yolov3-spp-sm2-1cls.cfg  # test
 gsutil cp -r backup/*.weights gs://sm4/weights  # weights to bucket
 
 # Debug/Development