Fix Common Errors on Demo/Configs

configs/attribute_predict_coarse/global_predictor_resnet_attr.py

@@ -16,14 +16,18 @@
         inter_channels=[2048, 4096],
         outchannels=4096),
     attr_predictor=dict(
-        type='AttrPredictor', inchannels=4096, outchannels=attribute_num),
-    loss_attr=dict(
-        type='BCEWithLogitsLoss',
-        ratio=1,
-        weight=None,
-        size_average=None,
-        reduce=None,
-        reduction='mean'),
+        type='AttrPredictor',
+        inchannels=4096,
+        outchannels=attribute_num,
+        loss_attr=dict(
+            type='BCEWithLogitsLoss',
+            ratio=1,
+            weight=None,
+            size_average=None,
+            reduce=None,
+            reduction='mean',
+        ),
+    ),
     pretrained='checkpoint/resnet50.pth')
 
 pooling = 'Global'

configs/attribute_predict_coarse/global_predictor_vgg_attr.py

@@ -15,14 +15,18 @@
         inter_channels=[512, 4096],
         outchannels=4096),
     attr_predictor=dict(
-        type='AttrPredictor', inchannels=4096, outchannels=attribute_num),
-    loss_attr=dict(
-        type='BCEWithLogitsLoss',
-        ratio=1,
-        weight=None,
-        size_average=None,
-        reduce=None,
-        reduction='mean'),
+        type='AttrPredictor',
+        inchannels=4096,
+        outchannels=attribute_num,
+        loss_attr=dict(
+            type='BCEWithLogitsLoss',
+            ratio=1,
+            weight=None,
+            size_average=None,
+            reduce=None,
+            reduction='mean',
+        ),
+    ),
     pretrained='checkpoint/vgg16.pth')
 
 pooling = 'Global'

configs/landmark_detect/landmark_detect_resnet.py

@@ -7,7 +7,7 @@
 
 model = dict(
     type='LandmarkDetector',
-    backbone=dict(type='ResNet', layer_setting='resnet50'),
+    backbone=dict(type='ResNet', setting='resnet50'),
     global_pool=dict(
         type='GlobalPooling',
         inplanes=(7, 7),

demo/test_attr_predictor.py

@@ -40,7 +40,8 @@ def main():
     img_tensor = get_img_tensor(args.input, args.use_cuda)
     # global attribute predictor will not use landmarks
     # just set a default value
-    landmark_tensor = torch.zeros(8)
+    # TODO: Add landmark demo support
+    landmark_tensor = torch.zeros(16).view(1, -1)
     cfg.model.pretrained = None
     model = build_predictor(cfg.model)
     load_checkpoint(model, args.checkpoint, map_location='cpu')

demo/test_cate_attr_predictor.py

@@ -41,7 +41,8 @@ def main():
     img_tensor = get_img_tensor(args.input, args.use_cuda)
     # global attribute predictor will not use landmarks
     # just set a default value
-    landmark_tensor = torch.zeros(8)
+    # TODO: Add landmark demo support
+    landmark_tensor = torch.zeros(16).view(1, -1)
 
     model = build_predictor(cfg.model)
     load_checkpoint(model, args.checkpoint, map_location='cpu')

demo/test_landmark_detector.py

@@ -45,7 +45,7 @@ def main():
 
     args = parse_args()
     cfg = Config.fromfile(args.config)
-
+    cfg.model.pretrained = None
     img_tensor, w, h = get_img_tensor(args.input, args.use_cuda, get_size=True)
 
     # build model and load checkpoint

mmfashion/utils/image.py

@@ -14,12 +14,13 @@ def get_img_tensor(img_path, use_cuda, get_size=False):
 
     img_size = (224, 224)  # crop image to (224, 224)
     img.thumbnail(img_size, Image.ANTIALIAS)
+    w, h = img.size
     img = img.convert('RGB')
     normalize = transforms.Normalize(
         mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
     transform = transforms.Compose([
-        transforms.RandomResizedCrop(img_size[0]),
-        transforms.RandomHorizontalFlip(),
+        transforms.Pad((max(h - w, 0)//2,
+                        max(w - h, 0)//2), padding_mode='edge'),
         transforms.ToTensor(),
         normalize,
     ])
@@ -62,8 +63,9 @@ def show_img(img_tensor):
 def draw_landmarks(img_file, landmarks, r=2):
     img = Image.open(img_file)
     draw = ImageDraw.Draw(img)
+    w, h = img.size
     for i, lm in enumerate(landmarks):
-        x = lm[0]
-        y = lm[1]
+        x = lm[0] * (w / 224.)
+        y = lm[1] * (h / 224.)
         draw.ellipse([(x - r, y - r), (x + r, y + r)], fill=(255, 0, 0, 0))
     img.show()