fix(detection): optimize dtype (#55)

official/vision/detection/layers/basic/functional.py

@@ -13,41 +13,6 @@ import megengine.functional as F
 from megengine.core import Tensor
 
 
-def get_padded_array_np(
-    array: np.ndarray, multiple_number: int = 32, pad_value: float = 0
-) -> np.ndarray:
-    """ pad the nd-array to multiple stride of th e
-
-    Args:
-        array (np.ndarray):
-            the array with the shape of [batch, channel, height, width]
-        multiple_number (int):
-            make the height and width can be divided by multiple_number
-        pad_value (int): the value to be padded
-
-    Returns:
-        padded_array (np.ndarray)
-    """
-    batch, chl, t_height, t_width = array.shape
-    padded_height = (
-        (t_height + multiple_number - 1) // multiple_number * multiple_number
-    )
-    padded_width = (t_width + multiple_number - 1) // multiple_number * multiple_number
-
-    padded_array = (
-        np.ones([batch, chl, padded_height, padded_width], dtype=np.float32) * pad_value
-    )
-
-    ndim = array.ndim
-    if ndim == 4:
-        padded_array[:, :, :t_height, :t_width] = array
-    elif ndim == 3:
-        padded_array[:, :t_height, :t_width] = array
-    else:
-        raise Exception("Not supported tensor dim: %d" % ndim)
-    return padded_array
-
-
 def get_padded_tensor(
     array: Tensor, multiple_number: int = 32, pad_value: float = 0
 ) -> Tensor:

official/vision/detection/layers/det/anchor.py

@@ -48,8 +48,8 @@ class DefaultAnchorGenerator(BaseAnchorGenerator):
     ):
         super().__init__()
         self.base_size = base_size
-        self.anchor_scales = np.array(anchor_scales)
-        self.anchor_ratios = np.array(anchor_ratios)
+        self.anchor_scales = np.array(anchor_scales, dtype=np.float32)
+        self.anchor_ratios = np.array(anchor_ratios, dtype=np.float32)
         self.offset = offset
 
     def _whctrs(self, anchor):

official/vision/detection/layers/det/box_utils.py

@@ -42,8 +42,8 @@ class BoxCoder(BoxCoderBase, metaclass=ABCMeta):
             reg_std(np.ndarray):  [x0_std, x1_std, y0_std, y1_std] or None
 
         """
-        self.reg_mean = np.array(reg_mean)[None, :]
-        self.reg_std = np.array(reg_std)[None, :]
+        self.reg_mean = np.array(reg_mean, dtype=np.float32)[None, :]
+        self.reg_std = np.array(reg_std, dtype=np.float32)[None, :]
         super().__init__()
 
     @staticmethod

official/vision/detection/models/faster_rcnn.py

@@ -68,8 +68,8 @@ class FasterRCNN(M.Module):
 
     def preprocess_image(self, image):
         normed_image = (
-            image - np.array(self.cfg.img_mean)[None, :, None, None]
-        ) / np.array(self.cfg.img_std)[None, :, None, None]
+            image - np.array(self.cfg.img_mean, dtype=np.float32)[None, :, None, None]
+        ) / np.array(self.cfg.img_std, dtype=np.float32)[None, :, None, None]
         return layers.get_padded_tensor(normed_image, 32, 0.0)
 
     def forward(self, inputs):

official/vision/detection/models/retinanet.py

@@ -33,7 +33,7 @@ class RetinaNet(M.Module):
         )
         self.box_coder = layers.BoxCoder(cfg.reg_mean, cfg.reg_std)
 
-        self.stride_list = np.array(cfg.stride).astype(np.float32)
+        self.stride_list = np.array(cfg.stride, dtype=np.float32)
         self.in_features = ["p3", "p4", "p5", "p6", "p7"]
 
         # ----------------------- build the backbone ------------------------ #
@@ -82,8 +82,8 @@ class RetinaNet(M.Module):
 
     def preprocess_image(self, image):
         normed_image = (
-            image - np.array(self.cfg.img_mean)[None, :, None, None]
-        ) / np.array(self.cfg.img_std)[None, :, None, None]
+            image - np.array(self.cfg.img_mean, dtype=np.float32)[None, :, None, None]
+        ) / np.array(self.cfg.img_std, dtype=np.float32)[None, :, None, None]
         return layers.get_padded_tensor(normed_image, 32, 0.0)
 
     def forward(self, inputs):

official/vision/detection/tools/test.py

@@ -64,7 +64,7 @@ def main():
             args.start_epoch = cfg.max_epoch - 1
         if args.end_epoch == -1:
             args.end_epoch = args.start_epoch
-    assert 0 <= args.start_epoch <= args.end_epoch < cfg.max_epoch
+        assert 0 <= args.start_epoch <= args.end_epoch < cfg.max_epoch
 
     for epoch_num in range(args.start_epoch, args.end_epoch + 1):
         if args.weight_file:
@@ -182,7 +182,7 @@ def worker(
         result_queue.put_nowait(
             {
                 "det_res": pred_res,
-                "image_id": int(data_dict[1][2][0].split(".")[0].split("_")[-1]),
+                "image_id": int(data_dict[1][2][0].split(".")[0]),
             }
         )