[MaskRCNN] update docs about training coco-format dataset (fix #1038)

examples/FasterRCNN/NOTES.md

@@ -16,10 +16,14 @@ This is a minimal implementation that simply contains these files:
 
 Data:
 
-1. It's easy to train on your own data. 
-   If your data is not in COCO format, you can just rewrite all the methods of
-   `DetectionDataset` following its documents in `dataset.py`.
-   You'll implement the logic to load your dataset and evaluate predictions.
+1. It's easy to train on your own data by changing `dataset.py`.
+
+   + If your data is in COCO format, modify `COCODetection`
+     to change the class names and the id mapping.
+   + If your data is not in COCO format, ignore `COCODetection` completely and
+     rewrite all the methods of
+     `DetectionDataset` following its documents.
+     You'll implement the logic to load your dataset and evaluate predictions.
 
 2. You can easily add more augmentations such as rotation, but be careful how a box should be
 	 augmented. The code now will always use the minimal axis-aligned bounding box of the 4 corners,

examples/FasterRCNN/dataset.py

@@ -25,8 +25,10 @@ class COCODetection(object):
     COCO_id_to_category_id = {1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9, 10: 10, 11: 11, 13: 12, 14: 13, 15: 14, 16: 15, 17: 16, 18: 17, 19: 18, 20: 19, 21: 20, 22: 21, 23: 22, 24: 23, 25: 24, 27: 25, 28: 26, 31: 27, 32: 28, 33: 29, 34: 30, 35: 31, 36: 32, 37: 33, 38: 34, 39: 35, 40: 36, 41: 37, 42: 38, 43: 39, 44: 40, 46: 41, 47: 42, 48: 43, 49: 44, 50: 45, 51: 46, 52: 47, 53: 48, 54: 49, 55: 50, 56: 51, 57: 52, 58: 53, 59: 54, 60: 55, 61: 56, 62: 57, 63: 58, 64: 59, 65: 60, 67: 61, 70: 62, 72: 63, 73: 64, 74: 65, 75: 66, 76: 67, 77: 68, 78: 69, 79: 70, 80: 71, 81: 72, 82: 73, 84: 74, 85: 75, 86: 76, 87: 77, 88: 78, 89: 79, 90: 80}  # noqa
     """
     Mapping from the incontinuous COCO category id to an id in [1, #category]
+    For your own dataset, this should usually be an identity mapping.
     """
 
+    # 80 names for COCO
     class_names = [
         "person", "bicycle", "car", "motorcycle", "airplane", "bus", "train", "truck", "boat", "traffic light", "fire hydrant", "stop sign", "parking meter", "bench", "bird", "cat", "dog", "horse", "sheep", "cow", "elephant", "bear", "zebra", "giraffe", "backpack", "umbrella", "handbag", "tie", "suitcase", "frisbee", "skis", "snowboard", "sports ball", "kite", "baseball bat", "baseball glove", "skateboard", "surfboard", "tennis racket", "bottle", "wine glass", "cup", "fork", "knife", "spoon", "bowl", "banana", "apple", "sandwich", "orange", "broccoli", "carrot", "hot dog", "pizza", "donut", "cake", "chair", "couch", "potted plant", "bed", "dining table", "toilet", "tv", "laptop", "mouse", "remote", "keyboard", "cell phone", "microwave", "oven", "toaster", "sink", "refrigerator", "book", "clock", "vase", "scissors", "teddy bear", "hair drier", "toothbrush"]  # noqa
 
@@ -191,10 +193,9 @@ class DetectionDataset(object):
         This function is responsible for setting the dataset-specific
         attributes in both cfg and self.
         """
-        self.num_category = cfg.DATA.NUM_CATEGORY = 80
+        self.num_category = cfg.DATA.NUM_CATEGORY = len(COCODetection.class_names)
         self.num_classes = self.num_category + 1
         self.class_names = cfg.DATA.CLASS_NAMES = ["BG"] + COCODetection.class_names
-        assert len(self.class_names) == self.num_classes
 
     def load_training_roidbs(self, names):
         """