misc small changes

Summary: on testing, typing

Reviewed By: rbgirshick

Differential Revision: D21158243

fbshipit-source-id: 84237575f3a65816b6988fa8994a8106853497e7

.circleci/config.yml

@@ -158,7 +158,7 @@ jobs:
       - run:
           name: Run Unit Tests
           command: |
-            docker exec -it d2 python3 -m unittest discover -v -s /detectron2/tests
+            docker exec -e CIRCLECI=true -it d2 python3 -m unittest discover -v -s /detectron2/tests
 
 workflows:
   version: 2

detectron2/checkpoint/detection_checkpoint.py

@@ -57,6 +57,8 @@ def _load_model(self, checkpoint):
             checkpoint["model"] = model_state_dict
         # for non-caffe2 models, use standard ways to load it
         incompatible = super()._load_model(checkpoint)
+        if incompatible is None:  # support older versions of fvcore
+            return None
 
         model_buffers = dict(self.model.named_buffers(recurse=False))
         for k in ["pixel_mean", "pixel_std"]:

detectron2/data/common.py

@@ -117,6 +117,8 @@ class AspectRatioGroupedDataset(data.IterableDataset):
     Batch data that have similar aspect ratio together.
     In this implementation, images whose aspect ratio < (or >) 1 will
     be batched together.
+    This improves training speed because the images then need less padding
+    to form a batch.
 
     It assumes the underlying dataset produces dicts with "width" and "height" keys.
     It will then produce a list of original dicts with length = batch_size,

detectron2/layers/wrappers.py

@@ -151,7 +151,7 @@ def forward(self, x):
             return _NewEmptyTensorOp.apply(x, output_shape)
 
 
-if False:  # not yet fixed in pytorch
+if TORCH_VERSION > (1, 5):
     Linear = torch.nn.Linear
 else:
 
@@ -182,7 +182,7 @@ def interpolate(input, size=None, scale_factor=None, mode="nearest", align_corne
     """
     A wrapper around :func:`torch.nn.functional.interpolate` to support zero-size tensor.
     """
-    if input.numel() > 0:
+    if TORCH_VERSION > (1, 4) or input.numel() > 0:
         return torch.nn.functional.interpolate(
             input, size, scale_factor, mode, align_corners=align_corners
         )

detectron2/modeling/anchor_generator.py

@@ -41,7 +41,7 @@ def __iter__(self):
         return iter(self._buffers.values())
 
 
-def _create_grid_offsets(size, stride, offset, device):
+def _create_grid_offsets(size: List[int], stride: int, offset: float, device: torch.device):
     grid_height, grid_width = size
     shifts_x = torch.arange(
         offset * stride, grid_width * stride, step=stride, dtype=torch.float32, device=device
@@ -154,7 +154,7 @@ def num_cell_anchors(self):
         """
         return [len(cell_anchors) for cell_anchors in self.cell_anchors]
 
-    def _grid_anchors(self, grid_sizes):
+    def _grid_anchors(self, grid_sizes: List[List[int]]):
         anchors = []
         for size, stride, base_anchors in zip(grid_sizes, self.strides, self.cell_anchors):
             shift_x, shift_y = _create_grid_offsets(size, stride, self.offset, base_anchors.device)

tests/data/test_detection_utils.py

@@ -4,7 +4,6 @@
 import numpy as np
 import unittest
 import pycocotools.mask as mask_util
-import torch
 
 from detectron2.data import detection_utils
 from detectron2.data import transforms as T

tests/modeling/test_anchor_generator.py

@@ -51,7 +51,6 @@ def test_default_anchor_generator_centered(self):
         # only the last two dimensions of features matter here
         num_images = 2
         features = {"stage3": torch.rand(num_images, 96, 1, 2)}
-        anchors = anchor_generator([features["stage3"]])
         expected_anchor_tensor = torch.tensor(
             [
                 [-30.0, -6.0, 34.0, 10.0],
@@ -69,9 +68,15 @@ def test_default_anchor_generator_centered(self):
             ]
         )
 
+        anchors = anchor_generator([features["stage3"]])
         for i in range(num_images):
             assert torch.allclose(anchors[i][0].tensor, expected_anchor_tensor)
 
+        # doesn't work yet
+        # anchors = torch.jit.script(anchor_generator)([features["stage3"]])
+        # for i in range(num_images):
+        #     assert torch.allclose(anchors[i][0].tensor, expected_anchor_tensor)
+
     def test_rrpn_anchor_generator(self):
         cfg = get_cfg()
         cfg.MODEL.ANCHOR_GENERATOR.SIZES = [[32, 64]]