update yolact, removing Instaboost when evaluating

yolact/README.md

@@ -17,15 +17,19 @@ pip install instaboost
 Integrating InstaBoost into yolact is quite simple.
 Only few lines should be modified to put InstaBoost into use.
 
-Only [data/coco.py](data/coco.py) should be changed.
+[data/coco.py](data/coco.py), [train.py](train.py) and [eval.py](eval.py) should be changed.
 
-[line 11](data/coco.py#L11) Importing InstaBoost.
+[coco.py/line 11](data/coco.py#L11) imports InstaBoost.
 
-[line 115-120](data/coco.py#L115-L120) These lines are originally located at [line 137-144](data/coco.py#L137-L144). They are put here because the InstaBoost function `get_new_data` needs both the annotation and the image.
+[coco.py/line 68](data/coco.py#L68) assigns the value of `is_train`.
 
-[line 122](data/coco.py#L122) Run the InstaBoost function and get new image and annotation. Both the variables `target` and `img` are acquired using the official COCO Python API.
+[coco.py/line 116-121](data/coco.py#L116-L121) These lines are originally located at [line 139-146](data/coco.py#L137-L144). They are put here because the InstaBoost function `get_new_data` needs both the annotation and the image.
 
+[coco.py/line 123-124](data/coco.py#L123-L124) checks if it's training or evaluating. When training, the program runs the InstaBoost function and gets new image and annotation. Both the variables `target` and `img` are acquired using the official COCO Python API.
 
+[train.py/line 130](train.py#L130) and [train.py/line 134](train.py#L134) assign `True` to the variable of `is_train`.
+
+[eval.py/line](eval.py#L) assigns `False` to the variable of `is_train`.
 ## Results and models
 
 For your conveinience of evaluation and comparison, we report the evaluation number on COCO val below. In our paper, the numbers are obtained from test-dev.

yolact/data/coco.py

@@ -61,10 +61,11 @@ class COCODetection(data.Dataset):
 
     def __init__(self, image_path, info_file, transform=None,
                  target_transform=COCOAnnotationTransform(),
-                 dataset_name='MS COCO', has_gt=True):
+                 dataset_name='MS COCO', has_gt=True,is_train=True):
         # Do this here because we have too many things named COCO
         from pycocotools.coco import COCO
 
+        self.is_train=is_train
         self.root = image_path
         self.coco = COCO(info_file)
 
@@ -95,7 +96,7 @@ def __len__(self):
     def pull_item(self, index):
         """
         Args:
-            index (int): Index
+            index (int): Indexcolor_cache
         Returns:
             tuple: Tuple (image, target, masks, height, width, crowd).
                    target is the object returned by ``coco.loadAnns``.
@@ -119,7 +120,8 @@ def pull_item(self, index):
         assert osp.exists(path), 'Image path does not exist: {}'.format(path)
         img = cv2.imread(path)
 
-        target, img = get_new_data(target, img, None, background=None)
+        if self.is_train:
+            target, img = get_new_data(target, img, None, background=None)
 
         # Separate out crowd annotations. These are annotations that signify a large crowd of
         # objects of said class, where there is no annotation for each individual object. Both

yolact/eval.py

@@ -1003,7 +1003,7 @@ def print_maps(all_maps):
 
         if args.image is None and args.video is None and args.images is None:
             dataset = COCODetection(cfg.dataset.valid_images, cfg.dataset.valid_info,
-                                    transform=BaseTransform(), has_gt=cfg.dataset.has_gt)
+                                    transform=BaseTransform(), has_gt=cfg.dataset.has_gt,is_train=False)
             prep_coco_cats()
         else:
             dataset = None        

yolact/train.py

@@ -131,13 +131,13 @@ def train():
 
     dataset = COCODetection(image_path=cfg.dataset.train_images,
                             info_file=cfg.dataset.train_info,
-                            transform=SSDAugmentation(MEANS))
+                            transform=SSDAugmentation(MEANS),is_train=True)
 
     if args.validation_epoch > 0:
         setup_eval()
         val_dataset = COCODetection(image_path=cfg.dataset.valid_images,
                                     info_file=cfg.dataset.valid_info,
-                                    transform=BaseTransform(MEANS))
+                                    transform=BaseTransform(MEANS),is_train=True)
 
     # Parallel wraps the underlying module, but when saving and loading we don't want that
     yolact_net = Yolact()