update docs

examples/FasterRCNN/coco.py

@@ -12,7 +12,6 @@ from tensorpack.utils.rect import FloatBox
 from tensorpack.utils.timer import timed_operation
 from tensorpack.utils.argtools import log_once
 
-from pycocotools.coco import COCO
 from config import config as cfg
 
 
@@ -63,6 +62,7 @@ class COCODetection(object):
             basedir, 'annotations/instances_{}.json'.format(name))
         assert os.path.isfile(annotation_file), annotation_file
 
+        from pycocotools.coco import COCO
         self.coco = COCO(annotation_file)
 
         # initialize the meta

examples/FasterRCNN/common.py

@@ -7,8 +7,6 @@ import cv2
 from tensorpack.dataflow import RNGDataFlow
 from tensorpack.dataflow.imgaug import transform
 
-import pycocotools.mask as cocomask
-
 
 class DataFromListOfDict(RNGDataFlow):
     def __init__(self, lst, keys, shuffle=False):
@@ -95,6 +93,8 @@ def segmentation_to_mask(polys, height, width):
         a binary matrix of (height, width)
     """
     polys = [p.flatten().tolist() for p in polys]
+
+    import pycocotools.mask as cocomask
     rles = cocomask.frPyObjects(polys, height, width)
     rle = cocomask.merge(rles)
     return cocomask.decode(rle)

examples/FasterRCNN/config.py

@@ -64,7 +64,7 @@ _C.DATA.BASEDIR = '/path/to/your/COCO/DIR'
 _C.DATA.TRAIN = ['train2014', 'valminusminival2014']   # i.e., trainval35k
 _C.DATA.VAL = 'minival2014'   # For now, only support evaluation on single dataset
 _C.DATA.NUM_CATEGORY = 80    # 80 categories.
-_C.DATA.CLASS_NAMES = []  # NUM_CLASS strings. Needs to be populated later by data loader
+_C.DATA.CLASS_NAMES = []  # NUM_CLASS (NUM_CATEGORY+1) strings, to be populated later by data loader. The first is BG.
 
 # basemodel ----------------------
 _C.BACKBONE.WEIGHTS = ''   # /path/to/weights.npz

examples/FasterRCNN/data.py

@@ -264,9 +264,9 @@ def get_train_dataflow():
     To train on your own data, change this to your loader.
     Produce "imgs" as a list of dict, in the dict the following keys are needed for training:
     height, width: integer
-    file_name: str
-    boxes: kx4 floats
-    class: k integers
+    file_name: str, full path to the image
+    boxes: numpy array of kx4 floats
+    class: numpy array of k integers
     is_crowd: k booleans. Use k False if you don't know what it means.
     segmentation: k lists of numpy arrays (one for each box).
         Each list of numpy array corresponds to the mask for one instance.

examples/FasterRCNN/model_frcnn.py

@@ -239,7 +239,7 @@ def fastrcnn_2fc_head(feature, num_classes):
 def fastrcnn_Xconv1fc_head(feature, num_classes, num_convs, norm=None):
     """
     Args:
-        feature (any shape):
+        feature (NCHW):
         num_classes(int): num_category + 1
         num_convs (int): number of conv layers
         norm (str or None): either None or 'GN'

examples/FasterRCNN/viz.py

@@ -12,6 +12,7 @@ from config import config as cfg
 
 
 def draw_annotation(img, boxes, klass, is_crowd=None):
+    """Will not modify img"""
     labels = []
     assert len(boxes) == len(klass)
     if is_crowd is not None:

tensorpack/tfutils/varmanip.py

@@ -178,7 +178,7 @@ def get_checkpoint_path(model_path):
     elif model_path.endswith('.index'):
         new_path = model_path.split('.index')[0]
     if new_path != model_path:
-        logger.warn(
+        logger.info(
             "Checkpoint path {} is auto-corrected to {}.".format(model_path, new_path))
         model_path = new_path
     assert tf.gfile.Exists(model_path) or tf.gfile.Exists(model_path + '.index'), model_path